1
00:00:01,001 --> 00:00:03,001
So I was looking for a thesis topic.

2
00:00:05,001 --> 00:00:07,001
And that's sometimes a hard thing.

3
00:00:07,001 --> 00:00:10,001
And that's sometimes why people take so
long to get their graduate school, to

4
00:00:10,001 --> 00:00:13,000
find what are you going to do?

5
00:00:13,000 --> 00:00:15,000
You only do that on thesis,
and it's kind of a big deal.

6
00:00:16,000 --> 00:00:18,001
And part of the problems with finding
a thesis topic is that I was being

7
00:00:18,001 --> 00:00:23,001
distracted by all sorts of things,
including ham radio, which I was getting more

8
00:00:23,001 --> 00:00:25,001
and more involved in the ham radio club.

9
00:00:28,000 --> 00:00:31,000
And what they were doing was
hidden transmitter hunting.

10
00:00:31,000 --> 00:00:35,000
And I was running around wasting time
looking for a hidden transmitter that

11
00:00:35,000 --> 00:00:36,001
kept getting hidden in the Boston area.

12
00:00:39,000 --> 00:00:42,000
And I was also getting
distracted by other fantasies.

13
00:00:43,000 --> 00:00:46,001
I was having technical fantasies to
build things and do neat things.

14
00:00:47,000 --> 00:00:50,001
And I was sort of in a computer science,
more oriented graduate program.

15
00:00:51,000 --> 00:00:54,000
And one of the things that always frustrated
me was the fact that I had to walk

16
00:00:54,000 --> 00:00:57,000
across campus to go back to freshman year.

17
00:00:57,000 --> 00:00:59,000
I had a 300-bit per second dial-up in my room.

18
00:01:00,000 --> 00:01:04,000
I'd have to walk across campus to get the
real useful connection to the computer

19
00:01:04,000 --> 00:01:08,000
in the net,
which was a 9,600-bit per second 2480 VT100.

20
00:01:09,000 --> 00:01:11,000
But by the time I was in graduate school,
I had the 9,600

21
00:01:11,000 --> 00:01:12,001
-bit per second modem in my room.

22
00:01:13,000 --> 00:01:16,000
But it wasn't useful because
I'd walk across campus.

23
00:01:16,001 --> 00:01:20,001
And by then I had an office where I had
a microvax2 running x-window system.

24
00:01:20,001 --> 00:01:21,001
And it was on the net.

25
00:01:22,001 --> 00:01:25,001
And on 10-megabit Ethernet,
well hooked to the internet.

26
00:01:25,001 --> 00:01:28,001
And frustration,
why can't I have the net at home?

27
00:01:29,000 --> 00:01:31,001
And so I'd always been
scheming on how to do this.

28
00:01:31,001 --> 00:01:35,001
Actually, one of the buildings I lived
in was the building which had the 10

29
00:01:35,001 --> 00:01:40,001
-megabit microwave for the near net,
which was the Boston area internet.

30
00:01:41,000 --> 00:01:42,001
And actually, I lived on the 8th floor.

31
00:01:43,000 --> 00:01:45,000
And up on the 26th floor was all the equipment.

32
00:01:45,000 --> 00:01:46,000
And they had routers and everything.

33
00:01:46,000 --> 00:01:52,000
And there was actually fiber, only six
feet from my room, where it was the trunk.

34
00:01:52,001 --> 00:01:52,001
And it's like.

35
00:01:57,000 --> 00:02:00,000
And I still had to dial in
with my 8600-bit per second.

36
00:02:00,000 --> 00:02:01,001
Frustration, frustration.

37
00:02:01,001 --> 00:02:04,001
This is a frustration I had,
I think, a better part of a decade.

38
00:02:05,001 --> 00:02:06,001
And so then I moved.

39
00:02:07,000 --> 00:02:07,001
And I looked.

40
00:02:07,001 --> 00:02:11,001
And the building they had my office in,
I could see

41
00:02:11,001 --> 00:02:13,000
many windows in that building.

42
00:02:13,000 --> 00:02:16,001
Not my own, but I could see many
windows in that building out my window.

43
00:02:17,001 --> 00:02:19,001
I don't remember the exact distance.

44
00:02:19,001 --> 00:02:20,001
There's one or two kilometers.

45
00:02:21,001 --> 00:02:22,001
I ought to be able to do something.

46
00:02:22,001 --> 00:02:24,000
And I started thinking free space optical

47
00:02:24,000 --> 00:02:26,000
links,
which actually there are a few of those around.

48
00:02:26,000 --> 00:02:28,000
But of course, they were very expensive

49
00:02:28,000 --> 00:02:29,001
for a graduate student budget in those days.

50
00:02:29,001 --> 00:02:31,000
So I started having other fantasies.

51
00:02:31,000 --> 00:02:33,000
Like maybe I could build a pair of radios.

52
00:02:33,000 --> 00:02:35,001
And actually, optics has a problem
that you have to get lenses.

53
00:02:35,001 --> 00:02:36,001
And you have to line it up.

54
00:02:36,001 --> 00:02:37,000
And it doesn't work when it rains.

55
00:02:38,000 --> 00:02:39,000
So I started thinking about radio.

56
00:02:39,000 --> 00:02:41,000
The problem with radio is,
well, you need a license.

57
00:02:41,000 --> 00:02:42,000
Well, I already have a hand license.

58
00:02:42,000 --> 00:02:47,001
But there's this content problem that
you can't just go read all your email,

59
00:02:48,000 --> 00:02:49,001
because you never know
what anybody has sent you.

60
00:02:50,000 --> 00:02:51,001
So you can't take responsibility for it.

61
00:02:52,000 --> 00:02:53,000
So I thought, well, I need something.

62
00:02:53,000 --> 00:02:56,000
And besides, I sort of wanted to
be able to let my friends do this.

63
00:02:56,001 --> 00:02:57,001
And most of them didn't have licenses.

64
00:02:57,001 --> 00:03:00,000
And so what I wanted was a loophole.

65
00:03:00,000 --> 00:03:07,000
So I actually went off and
I sat down in the small

66
00:03:07,000 --> 00:03:09,001
law library at the business school at MIT.

67
00:03:10,000 --> 00:03:16,001
And I actually sat down and I started
reading 47 CFR in general, and particularly

68
00:03:16,001 --> 00:03:21,000
47 CFR 15, which is the place I
expected I'd find a loophole.

69
00:03:21,000 --> 00:03:22,001
That's what people call 15.

70
00:03:22,001 --> 00:03:25,001
This is sort of what a
lawyer's would call part 10.

71
00:03:26,000 --> 00:03:30,001
This is volume 47 of the Code of Federal
Regulations, which are the regulations

72
00:03:30,001 --> 00:03:32,001
that are written by the bureaucrats,
not the laws that are made

73
00:03:32,001 --> 00:03:34,001
by the legislative process.

74
00:03:35,001 --> 00:03:40,000
Anyway, this is with the part 15 unlicensed
authorization that I discovered.

75
00:03:40,001 --> 00:03:42,000
And this was probably late in 91.

76
00:03:44,000 --> 00:03:46,001
I'm reading along, looking at each
of the things you can do in part 15.

77
00:03:46,001 --> 00:03:51,000
And I'm like, wow,
you can do a watt as long as your

78
00:03:51,000 --> 00:03:52,001
antenna gain is at most 6 dVI.

79
00:03:53,000 --> 00:03:58,001
So sort of an arp of 4 watts in three
different bands, the widest of which is

80
00:03:58,001 --> 00:04:00,000
125 megahertz wide.

81
00:04:01,000 --> 00:04:06,000
But this regulation required that you
use spread spectrum with at least 10 dB

82
00:04:06,000 --> 00:04:07,001
of processing gain in the receiver.

83
00:04:10,000 --> 00:04:14,000
Most people are ignoring this requirement
that you actually demonstrate the

84
00:04:14,000 --> 00:04:16,001
system of 10 dB of processing
gain in the receiver these days.

85
00:04:17,001 --> 00:04:21,000
But anyway,
sitting there reading the FCC regulations is

86
00:04:21,000 --> 00:04:23,001
like, what's the spread spectrum?

87
00:04:24,000 --> 00:04:25,001
And what's the processing gain?

88
00:04:26,001 --> 00:04:31,001
And I had sort of gone through the computer
science curriculum, so I hadn't had

89
00:04:31,001 --> 00:04:34,001
all the modulation and that sort of theory.

90
00:04:35,000 --> 00:04:38,001
So that sent me off studying,
looking in the library.

91
00:04:40,000 --> 00:04:42,000
And I started thinking, oh, yeah,
that's something I could do.

92
00:04:42,000 --> 00:04:47,000
And I started having fantasies of building up,
here's me.

93
00:04:47,001 --> 00:04:48,001
Here's the building I want to get to.

94
00:04:51,000 --> 00:04:51,001
I want that.

95
00:04:55,001 --> 00:04:57,000
And that'd be neat.

96
00:04:57,000 --> 00:05:01,000
And of course, any time you go to the
trouble of spinning some boards and

97
00:05:01,000 --> 00:05:05,001
populating them,
you don't build just two boards.

98
00:05:06,000 --> 00:05:10,000
You actually get a dozen boards
made and maybe a prototype.

99
00:05:10,001 --> 00:05:12,000
And then you figure out
how to really make them.

100
00:05:12,000 --> 00:05:13,000
And then you sell them to your friends.

101
00:05:14,000 --> 00:05:15,000
And I had all these friends in grad school.

102
00:05:15,000 --> 00:05:17,001
And I thought, they're all over the place.

103
00:05:17,001 --> 00:05:19,000
And they're all going to want this, too.

104
00:05:19,001 --> 00:05:21,000
They're just like me, sort of.

105
00:05:21,001 --> 00:05:25,000
And I started thinking, yeah,
this would be really neat.

106
00:05:28,001 --> 00:05:33,000
But the internet wasn't just a,
by this point, the internet was not just

107
00:05:33,000 --> 00:05:37,001
something that all the computer
science students were addicted to.

108
00:05:37,001 --> 00:05:41,000
Even the biology graduate students
were addicted to the internet.

109
00:05:41,000 --> 00:05:43,000
They couldn't go a day
without reading their email.

110
00:05:45,000 --> 00:05:47,001
And they probably wanted it home, as well.

111
00:05:47,001 --> 00:05:49,000
This is sort of before the web.

112
00:05:50,001 --> 00:05:52,000
Or the web was just getting started then,
anyway.

113
00:05:53,000 --> 00:05:56,001
So I realized it's not going
to necessarily just be this.

114
00:05:58,000 --> 00:06:03,000
It might be other groups around
the campus or around the area

115
00:06:03,000 --> 00:06:04,001
might want to do this, as well.

116
00:06:04,001 --> 00:06:06,000
And it's like, oh, that's interesting.

117
00:06:07,001 --> 00:06:10,000
Well, maybe their community looks like that.

118
00:06:10,000 --> 00:06:12,001
There's this little network
of me and my friends here.

119
00:06:12,001 --> 00:06:14,000
But maybe everybody else distributed and

120
00:06:14,000 --> 00:06:16,000
tried to use the same things to get here.

121
00:06:16,000 --> 00:06:21,000
And I'm having fantasies of selling these
things as kits or maybe even a product.

122
00:06:22,001 --> 00:06:23,000
Full of fantasy.

123
00:06:23,001 --> 00:06:28,000
But then I spent a lot of time just
thinking about this sort of stuff.

124
00:06:28,000 --> 00:06:32,001
And I realized that it's probably not right
to have a hub system, but to actually

125
00:06:32,001 --> 00:06:34,001
just view it as you've got all these stations.

126
00:06:35,000 --> 00:06:38,000
And what you really want to do is get
them all connected together in a network.

127
00:06:38,001 --> 00:06:40,001
And I was also thinking at the time.

128
00:06:40,001 --> 00:06:46,000
And I realized that you need to
not hop over intermediate stations

129
00:06:46,000 --> 00:06:47,001
or the system cannot scale.

130
00:06:48,001 --> 00:06:52,001
And that was sort of the beginning of
me thinking about how can this scale.

131
00:06:52,001 --> 00:06:56,000
And then I actually realized that, oh,
it's not just the biology graduate

132
00:06:56,000 --> 00:06:57,001
students who want to get
on the internet as well.

133
00:06:58,000 --> 00:07:00,000
It's also the entire population.

134
00:07:00,001 --> 00:07:06,001
At that point, I think I sort of realized
that the internet is something that

135
00:07:06,001 --> 00:07:10,000
eventually the whole population is
going to be on one way or another.

136
00:07:12,000 --> 00:07:14,000
And being behind a modem is not really on the

137
00:07:14,000 --> 00:07:17,000
net, at least from my point of view.

138
00:07:17,000 --> 00:07:19,001
So I started thinking about,
well, how can this scale?

139
00:07:20,000 --> 00:07:23,001
To maybe 100 stations or maybe 1,000 stations.

140
00:07:24,001 --> 00:07:25,001
And then I thought, oh, no.

141
00:07:26,000 --> 00:07:29,000
Well, if actually how many houses
are there in the Metro Boston area?

142
00:07:30,001 --> 00:07:31,001
Or Metro Beijing?

143
00:07:33,000 --> 00:07:34,000
Or whatever.

144
00:07:34,000 --> 00:07:39,001
And I was thinking about, I was reading
packet radio research reading propagation

145
00:07:39,001 --> 00:07:43,001
stuff in the library, trying to understand,
well, what is the propagation model?

146
00:07:45,000 --> 00:07:47,000
And how can you design a packet radio stuff?

147
00:07:47,000 --> 00:07:50,000
And by this time, I'm reading some of the
real packet radio research that's been

148
00:07:50,000 --> 00:07:52,000
funded by DARPA starting around 1970.

149
00:07:52,000 --> 00:07:54,001
There's about 25 years worth
of literature in the library.

150
00:07:55,000 --> 00:07:56,001
You can actually spend the time
to go look through it all.

151
00:07:57,001 --> 00:07:59,000
I was trying to understand it.

152
00:08:00,001 --> 00:08:03,000
A lot of the packet radio research model,
their model

153
00:08:03,000 --> 00:08:05,000
is done by computer scientists.

154
00:08:05,000 --> 00:08:10,001
And their model of a radio system
is a transmitter and a receiver.

155
00:08:10,001 --> 00:08:11,001
And it has some range.

156
00:08:12,000 --> 00:08:13,001
And if you're inside that range, it works.

157
00:08:13,001 --> 00:08:15,000
And if you're outside that range,
it doesn't work.

158
00:08:16,001 --> 00:08:19,000
And if there's another station transmitting,
well, if you're inside

159
00:08:19,000 --> 00:08:20,001
his range, then he interferes with you.

160
00:08:20,001 --> 00:08:23,000
And if you're outside his range,
then he doesn't interfere with you.

161
00:08:23,000 --> 00:08:26,000
And sort of a very determined
transmission radius.

162
00:08:27,000 --> 00:08:30,001
And you can actually go find papers
talking about, well, what is the right

163
00:08:30,001 --> 00:08:32,001
transmission range to set your transmitter to?

164
00:08:33,001 --> 00:08:35,001
To let the packet radio scale.

165
00:08:37,000 --> 00:08:39,001
And this wasn't making too much sense to me,
because I was out

166
00:08:39,001 --> 00:08:41,000
hidden transmitter hunting at the time.

167
00:08:43,000 --> 00:08:45,000
And Radio Prop Free Space is 1 over r squared.

168
00:08:45,001 --> 00:08:47,001
And if you go behind a hill,
it's different stuff.

169
00:08:49,000 --> 00:08:51,001
Some of the more serious land
mobile propagation stuff

170
00:08:51,001 --> 00:08:53,000
talked about 1 over r to the fourth.

171
00:08:55,000 --> 00:08:59,001
And I'm just now starting to understand
where some of these numbers come from.

172
00:08:59,001 --> 00:09:06,000
But then actually, about this time,
I went to Albuquerque New Mexico for some of

173
00:09:06,000 --> 00:09:07,001
my sisters-in-law's big Thanksgiving gathering.

174
00:09:08,000 --> 00:09:10,000
And I was visiting them for
Thanksgiving that year.

175
00:09:10,001 --> 00:09:14,000
And I arrived in Albuquerque late at night,
was shown up to an upstairs room,

176
00:09:14,000 --> 00:09:17,000
looked out the window the
next morning when I woke up.

177
00:09:17,000 --> 00:09:20,001
And I believed I could see every rooftop in

178
00:09:20,001 --> 00:09:22,001
Albuquerque, New Mexico, from where I was.

179
00:09:23,001 --> 00:09:27,001
And it occurred to me that, at least in
my naive understanding at the time, that

180
00:09:27,001 --> 00:09:32,000
it would probably be effectively
free-space path loss between all pairs of

181
00:09:32,000 --> 00:09:34,000
rooftops in that area.

182
00:09:34,001 --> 00:09:39,000
And that that is the most challenging
to scaling, is if it is free-space.

183
00:09:39,001 --> 00:09:42,000
Because free-space is 1 over r squared.

184
00:09:42,000 --> 00:09:45,000
And most of the other models indicate
that you get more attenuation than that.

185
00:09:45,001 --> 00:09:51,001
And I realized that if I can figure out
how to make it scale in the free-space

186
00:09:51,001 --> 00:09:54,001
case, then it can scale in any other
propagation model that anybody

187
00:09:54,001 --> 00:09:56,001
thinks is reasonable.

188
00:09:57,000 --> 00:10:01,000
The free-space propagation is
the hard case for scaling.

189
00:10:01,001 --> 00:10:03,000
So that's the one to solve.

190
00:10:05,001 --> 00:10:07,000
So at that point, I was off.

191
00:10:07,000 --> 00:10:09,000
And I was actually thinking,
well, how about a million

192
00:10:09,000 --> 00:10:10,001
stations in a metropolitan area?

193
00:10:13,000 --> 00:10:18,000
And packet radio networks,
all the stations are the same.

194
00:10:18,000 --> 00:10:21,000
It's not like a cellular system where
you've got a base station in the thing.

195
00:10:21,000 --> 00:10:25,001
And actually, I don't really, I mean,
I'm thinking of a fixed packet radio

196
00:10:25,001 --> 00:10:27,000
network where all the stations are fixed.

197
00:10:28,000 --> 00:10:29,001
Maybe I can figure out how
to add mobility later.

198
00:10:29,001 --> 00:10:30,001
But I didn't really care about that.

199
00:10:30,001 --> 00:10:32,000
I just wanted to get the
internet out to the houses.

200
00:10:33,000 --> 00:10:36,000
And so by base stations, I mean no cell sites.

201
00:10:37,000 --> 00:10:37,001
They're sort of all base stations.

202
00:10:38,000 --> 00:10:38,001
They're all peers.

203
00:10:38,001 --> 00:10:39,000
They're all the same.

204
00:10:39,000 --> 00:10:40,001
They form this mesh-not work.

205
00:10:41,000 --> 00:10:42,001
It needs to be multi-hop routing.

206
00:10:43,000 --> 00:10:46,000
I had some intuition to that at the
time that I really believed in.

207
00:10:46,001 --> 00:10:48,000
And I want a single-channel system.

208
00:10:49,001 --> 00:10:52,000
These stations are just people by them.

209
00:10:52,000 --> 00:10:53,000
They put them on top of their house.

210
00:10:54,001 --> 00:10:56,000
And they don't have to point them.

211
00:10:56,000 --> 00:10:59,001
They don't have to think about
whatever that should just work.

212
00:11:00,000 --> 00:11:01,001
So I was thinking on the antennas.

213
00:11:06,000 --> 00:11:12,001
And I wanted to make it scale to millions
of stations in the metro area while

214
00:11:12,001 --> 00:11:14,001
maintaining reasonable per user

215
00:11:14,001 --> 00:11:17,000
performance,
whatever those traffic they're doing.

216
00:11:17,000 --> 00:11:19,001
And I don't really know what the
traffic load on this net's going to be.

217
00:11:19,001 --> 00:11:21,000
I don't know who's going to talk to who.

218
00:11:21,001 --> 00:11:23,000
So it needs to be sort of
a general purpose net.

219
00:11:24,001 --> 00:11:27,000
Kind of like when you install an ethernet,
you usually don't worry

220
00:11:27,000 --> 00:11:28,001
about what the traffic matrix is going to be.

221
00:11:28,001 --> 00:11:29,001
You just kind of string it everywhere.

222
00:11:30,000 --> 00:11:32,000
And if it gets too busy,
then you start thinking about it.

223
00:11:34,001 --> 00:11:38,001
So the real concern is my original
goal was just to get my one link.

224
00:11:40,000 --> 00:11:42,001
But if there were millions of these,
is mine still going to work?

225
00:11:44,000 --> 00:11:47,000
Maybe actually there's some point to
where it falls over as you scale it.

226
00:11:47,001 --> 00:11:49,001
And my question is,
is there a way to design these

227
00:11:49,001 --> 00:11:51,001
things so that it all still works?

228
00:11:52,000 --> 00:11:57,001
It works with my two, and then it works
after a million more have been sold.

229
00:12:02,000 --> 00:12:03,001
So this is all thinking.

230
00:12:04,001 --> 00:12:08,000
At some point along the way, I realized,
oh, this is my thesis topic.

231
00:12:08,000 --> 00:12:11,001
So I actually wound up talking
a committee into actually

232
00:12:11,001 --> 00:12:13,000
letting me do this as a thesis topic.

233
00:12:15,000 --> 00:12:15,001
What?

234
00:12:16,000 --> 00:12:16,001
The ECS department?

235
00:12:17,000 --> 00:12:19,001
Well, it's an ECS department.

236
00:12:19,001 --> 00:12:25,001
[...] and actually, if you think about
who is computer science faculty,

237
00:12:26,000 --> 00:12:29,001
well, they were double the graduate
students because there was no computer

238
00:12:29,001 --> 00:12:31,001
science when they went through grad school.

239
00:12:32,000 --> 00:12:33,000
They kind of remember this stuff.

240
00:12:33,001 --> 00:12:35,001
I taught my particular supervisor
didn't know much about

241
00:12:35,001 --> 00:12:37,000
radio except the courses he'd taken.

242
00:12:38,000 --> 00:12:42,000
And I wound up teaching him a lot,
which is kind of neat about radios.

243
00:12:42,000 --> 00:12:43,001
He knows a lot about systems.

244
00:12:44,000 --> 00:12:49,001
Anyway, so I think there are two important
things that come out of the thesis.

245
00:12:49,001 --> 00:12:53,000
And this is the most important thing.

246
00:12:53,001 --> 00:12:55,001
And actually, I'm going to be
talking about this pretty soon here.

247
00:12:57,000 --> 00:13:04,000
And then I've got a way of scheduling
the packets so that you can

248
00:13:04,000 --> 00:13:07,001
send the packet to your nearby
neighbor without having

249
00:13:07,001 --> 00:13:09,000
to worry about whether or not it got there.

250
00:13:09,001 --> 00:13:13,001
So it'd be nice if you're going across
a thousand station network per hop

251
00:13:13,001 --> 00:13:15,001
acknowledgment is going to be a lot of stuff.

252
00:13:15,001 --> 00:13:17,001
It'd be nice if you could just use end-to-end

253
00:13:17,001 --> 00:13:19,001
acknowledgments across the whole network.

254
00:13:19,001 --> 00:13:20,001
Can you say those words I keep?

255
00:13:22,001 --> 00:13:28,001
Oh, this says understanding of noise
growth of a large system and decentralized

256
00:13:28,001 --> 00:13:30,001
collision-free channel access method.

257
00:13:33,001 --> 00:13:35,001
So this understanding of
the noise growth of a large

258
00:13:35,001 --> 00:13:37,001
system that I think is
the most important thing.

259
00:13:38,000 --> 00:13:41,000
I've come up with a particular way
of doing the decentralized collision

260
00:13:41,000 --> 00:13:42,001
-free channel access.

261
00:13:42,001 --> 00:13:46,001
And that MIT filed for a patent on.

262
00:13:48,000 --> 00:13:50,000
And I have not yet figured
out how to invent around it.

263
00:13:50,000 --> 00:13:54,001
But actually,
I think math paper yesterday suggests to me

264
00:13:54,001 --> 00:13:56,000
that actually there may be a way.

265
00:13:59,000 --> 00:14:00,000
Anyway, onward forward.

266
00:14:00,001 --> 00:14:02,000
OK, now how to understand.

267
00:14:02,001 --> 00:14:03,000
Oh, question.

268
00:14:03,001 --> 00:14:04,000
You mentioned that.

269
00:14:04,000 --> 00:14:07,000
Did you indicate to MIT that the graduate
student at MIT shared the patents?

270
00:14:08,000 --> 00:14:11,000
MIT owns the rights to the patents and all

271
00:14:11,000 --> 00:14:12,001
decisions about licensing and everything.

272
00:14:12,001 --> 00:14:17,000
And so some sense I have no control that
if they ever actually do manage to make

273
00:14:17,000 --> 00:14:20,001
money off of it,
which is not likely in general.

274
00:14:21,000 --> 00:14:23,001
Only one in 10 patents ever gets a license.

275
00:14:26,001 --> 00:14:29,001
They chose to go ahead and do
it because they thought it was

276
00:14:29,001 --> 00:14:31,000
worth the gamble given what it was.

277
00:14:31,000 --> 00:14:36,000
But I think I get some 10% or 15%
of whatever license fees they bring in.

278
00:14:38,000 --> 00:14:41,000
I'm not ever expecting
that to produce any money.

279
00:14:42,001 --> 00:14:44,000
Patents aren't really about producing money.

280
00:14:44,001 --> 00:14:49,000
They're this legal game that
companies play chicken game.

281
00:14:49,001 --> 00:14:53,000
Companies play chicken with each
other about perhaps going into

282
00:14:53,000 --> 00:14:54,001
court and whatever.

283
00:14:55,000 --> 00:14:56,000
But it was kind of neat.

284
00:14:57,000 --> 00:15:00,000
I got to see what it's like
to apply for a patent.

285
00:15:00,000 --> 00:15:01,001
I mean, they hired a patent attorney and stuff.

286
00:15:02,000 --> 00:15:04,000
I didn't really want to- [...] a name.

287
00:15:04,001 --> 00:15:05,000
Yeah.

288
00:15:06,000 --> 00:15:07,001
Yeah, it was written by the lawyer.

289
00:15:07,001 --> 00:15:09,001
The lawyer read my thesis and understood it.

290
00:15:09,001 --> 00:15:10,001
I was very impressed.

291
00:15:11,001 --> 00:15:14,000
He had been a double e undergrad
before he went to law school.

292
00:15:18,001 --> 00:15:19,000
So OK.

293
00:15:20,000 --> 00:15:27,000
My first concern is what happens as the thing

294
00:15:27,000 --> 00:15:30,000
scales to the radio channel if these
are all one single channel system.

295
00:15:34,000 --> 00:15:34,001
I'm here.

296
00:15:35,000 --> 00:15:37,000
I've got a nearby neighbor that's
trying to send me something.

297
00:15:37,001 --> 00:15:38,001
There's some signal there.

298
00:15:39,001 --> 00:15:43,001
And there's a whole bunch of other
stations out there using the same channel.

299
00:15:45,000 --> 00:15:45,001
Does it work?

300
00:15:49,001 --> 00:15:56,001
And actually, I was trying to place

301
00:15:56,001 --> 00:15:59,000
this in the context of the
packet radio literature.

302
00:15:59,000 --> 00:16:04,000
If I'm talking to a computer scientist
who's read some of this packet radio

303
00:16:04,000 --> 00:16:10,000
literature and it's published, I need to
explain what I'm doing that's different.

304
00:16:10,001 --> 00:16:13,001
And in those systems, the signal is
something that's either there or not.

305
00:16:14,001 --> 00:16:15,001
And transmitters have a range.

306
00:16:16,000 --> 00:16:19,000
And they design on this
transmission radius thing.

307
00:16:19,000 --> 00:16:23,000
So just in general,
you can look at any of these models

308
00:16:23,000 --> 00:16:25,000
for what a packet radio network is like.

309
00:16:25,001 --> 00:16:30,001
They've got some idea of what is a
signal and what is propagation and what

310
00:16:30,001 --> 00:16:36,000
criteria are there on successfully
capturing a packet that's been sent to you.

311
00:16:36,001 --> 00:16:40,000
And other systems have other
answers to those three questions.

312
00:16:40,000 --> 00:16:45,001
But in my case, the signal at the
receiver has a power level, which

313
00:16:45,001 --> 00:16:47,000
is a function of a whole bunch of stuff.

314
00:16:47,001 --> 00:16:48,001
It's a continuum.

315
00:16:49,000 --> 00:16:54,001
And radio signals propagate to the end of
the universe with some free space path

316
00:16:54,001 --> 00:16:58,000
loss or perhaps the attenuating
media in between or whatever.

317
00:16:59,000 --> 00:17:01,000
And by propagation, there's an attenuation.

318
00:17:02,001 --> 00:17:05,001
It's the model I use in all the work I did.

319
00:17:06,001 --> 00:17:13,000
Actually, in general, if you have two
antennas in the world and you assume that

320
00:17:13,000 --> 00:17:19,001
the world between the antennas is
linear and time invariant, then

321
00:17:20,001 --> 00:17:25,000
with those two assumptions,
then the most general model of what's between

322
00:17:25,000 --> 00:17:26,001
the antennas is an impulse response.

323
00:17:27,001 --> 00:17:30,001
That captures all of the
multipath between those antennas.

324
00:17:31,001 --> 00:17:34,001
And then you could do the Fourier
transform of that and understand what

325
00:17:34,001 --> 00:17:36,001
is the filter between the two antennas.

326
00:17:37,000 --> 00:17:39,001
And it's one with a lot of
attenuation and everything.

327
00:17:40,001 --> 00:17:44,000
And what I did is I dropped the impulse
response and said, well, I'm going to

328
00:17:44,000 --> 00:17:48,000
ignore the multipath for now because
in a multipath, you can design

329
00:17:48,000 --> 00:17:49,001
modulation schemes around that.

330
00:17:52,001 --> 00:17:57,001
And any problem that the multipath gives
you is probably just going to be you can

331
00:17:57,001 --> 00:17:59,000
design around it.

332
00:17:59,000 --> 00:18:01,000
I wanted to wave my hands
here and get beyond that.

333
00:18:01,000 --> 00:18:03,000
But you can design around that.

334
00:18:03,000 --> 00:18:06,001
And it affects you by some small factor
in how many bits you can get through.

335
00:18:08,000 --> 00:18:13,001
So for the purposes of understanding
noise growth system, between every

336
00:18:13,001 --> 00:18:15,000
pair of stations, there's some attenuation.

337
00:18:15,000 --> 00:18:18,000
It's the path loss between those two things.

338
00:18:18,000 --> 00:18:24,001
And it's actually I think
of this as a path gain,

339
00:18:25,001 --> 00:18:27,001
which is 1 over the path loss.

340
00:18:29,000 --> 00:18:30,000
So it's a small number.

341
00:18:30,000 --> 00:18:34,001
So you take the power at the station
you're transmitting ascending.

342
00:18:36,000 --> 00:18:42,001
And this term here is this
is the power at station k.

343
00:18:43,000 --> 00:18:47,001
And this term here is the attenuation
between the two stations.

344
00:18:48,000 --> 00:18:49,001
It's probably symmetric.

345
00:18:51,000 --> 00:18:53,001
And so you multiply that times
the power they're sending.

346
00:18:53,001 --> 00:18:56,000
And that tells you how much signal
you're getting from that station.

347
00:18:56,001 --> 00:19:01,000
And then for all the other stations,
in fact, this term here is to sum up the

348
00:19:01,000 --> 00:19:07,000
power that everybody else is
transmitting times how much

349
00:19:07,000 --> 00:19:09,000
attenuation you're getting from them.

350
00:19:09,001 --> 00:19:14,000
And if I just did this over this,
it would be signal to signal plus noise

351
00:19:14,000 --> 00:19:15,001
ratio in power.

352
00:19:17,000 --> 00:19:21,001
And if I subtract out this term,
then I get signal to noise.

353
00:19:22,000 --> 00:19:23,000
Because this is the signal.

354
00:19:23,001 --> 00:19:26,000
This term includes the signal and the noise.

355
00:19:26,000 --> 00:19:28,000
And I'm treating all of the other,
what's coming

356
00:19:28,000 --> 00:19:29,001
from all the other stations, as noise.

357
00:19:30,000 --> 00:19:31,000
I'm lumping that in.

358
00:19:31,000 --> 00:19:34,000
Sometimes people say signal
to noise plus interference.

359
00:19:34,001 --> 00:19:36,000
And they actually put an explicit i over there.

360
00:19:36,000 --> 00:19:38,000
I'm lumping all of the interference into noise.

361
00:19:39,001 --> 00:19:42,000
Because this is sort of going
to be spread spectrum like.

362
00:19:42,000 --> 00:19:43,001
And maybe everybody's using random codes.

363
00:19:46,001 --> 00:19:50,001
So by propagation, what you can understand that

364
00:19:50,001 --> 00:19:52,001
there's propagation from every station.

365
00:19:53,000 --> 00:19:56,000
And what you care about in order to
understand whether you've gotten a successful

366
00:19:56,000 --> 00:19:59,001
reception is whether the signal
to noise ratio is sufficient.

367
00:20:00,000 --> 00:20:04,001
And sort of my criteria for successful
reception, what I've done is I've taken

368
00:20:04,001 --> 00:20:09,000
Shannon's capacity theorem and
solved it for signal to noise ratio.

369
00:20:10,000 --> 00:20:12,000
Remember your algebra, you can do that.

370
00:20:12,001 --> 00:20:16,000
And I've thrown in, and I chose a log base 2 so

371
00:20:16,000 --> 00:20:17,001
that I'm measuring information in bits.

372
00:20:18,001 --> 00:20:22,000
And then say you have the bit rate you
want to communicate at and the bandwidth.

373
00:20:22,001 --> 00:20:24,001
And here it's 2 to that power.

374
00:20:25,000 --> 00:20:26,000
Subtract 1.

375
00:20:26,000 --> 00:20:30,001
And then it's got a fudge factor here,
which is all of that.

376
00:20:31,000 --> 00:20:35,001
This is my abstraction of a radio engineer,
a good communication engineer.

377
00:20:36,000 --> 00:20:37,000
You need the fudge factor.

378
00:20:37,000 --> 00:20:41,000
And probably needs a fudge factor
of somewhere between 3 and 10.

379
00:20:41,001 --> 00:20:43,001
Or maybe with fancy turbo code,
the only needs a

380
00:20:43,001 --> 00:20:45,000
fudge factor of 1.5 or something.

381
00:20:46,000 --> 00:20:49,001
And actually, I'm waving my hands a little
bit here because when you're sending

382
00:20:49,001 --> 00:20:52,000
packets, you have to synchronize.

383
00:20:52,000 --> 00:20:55,000
And so how do you build a burst modem that
actually approaches the Shannon limit?

384
00:20:55,000 --> 00:20:55,001
Well, that's tough.

385
00:20:57,001 --> 00:21:00,001
My answer to that is, well, actually,
it doesn't have to just be a burst modem

386
00:21:00,001 --> 00:21:06,000
because if you've been swapping packets
with this other station multiple times

387
00:21:06,000 --> 00:21:11,000
per second, then perhaps you can
maintain some timing information and not

388
00:21:11,000 --> 00:21:13,000
actually have to start from scratch on every

389
00:21:13,000 --> 00:21:14,001
packet to get everything synchronized up.

390
00:21:16,001 --> 00:21:20,000
So actually, I think some of the commercial
Part 15 radios that give you point-to

391
00:21:20,000 --> 00:21:22,001
-point data links, the radios go like this.

392
00:21:23,000 --> 00:21:29,001
And they actually maintain spreading
code synchronization across the, yeah,

393
00:21:30,000 --> 00:21:31,001
so they don't have to actually go hunt.

394
00:21:31,001 --> 00:21:34,001
And so they're not really
trying to build burst modems.

395
00:21:35,000 --> 00:21:38,000
But anyway, so this is my criteria
for successful reception.

396
00:21:42,000 --> 00:21:44,000
Now, what about scaling?

397
00:21:46,000 --> 00:21:50,001
The concern is here that my
neighbor is here near me.

398
00:21:51,000 --> 00:21:53,000
And I want to hear him.

399
00:21:53,000 --> 00:21:54,001
And I want to decode his packet.

400
00:21:55,001 --> 00:21:58,001
And whether or not I'm going to
be able to do that depends on how

401
00:21:58,001 --> 00:22:00,000
much other stuff is out there.

402
00:22:01,001 --> 00:22:04,001
And that depends on how far we've scaled.

403
00:22:05,000 --> 00:22:11,000
And the most general thing to do is,
well, what if... there's an

404
00:22:11,000 --> 00:22:13,000
unlimited number of stations out there.

405
00:22:13,001 --> 00:22:16,000
And I started thinking about that.

406
00:22:16,001 --> 00:22:18,001
And I started thinking about that problem.

407
00:22:18,001 --> 00:22:21,000
I had some friends who were like
physicists and astronomers.

408
00:22:22,000 --> 00:22:26,000
And I went rambling to them, and they said,
oh, that's just like Olver's Paradox.

409
00:22:26,001 --> 00:22:30,000
And Olver's Paradox is something in astronomy.

410
00:22:32,000 --> 00:22:32,001
Here's a question.

411
00:22:33,000 --> 00:22:34,000
How bright is the sky at night?

412
00:22:34,001 --> 00:22:36,000
Well, we can figure this out.

413
00:22:37,000 --> 00:22:40,001
We live in an infinitely old universe, perhaps.

414
00:22:41,001 --> 00:22:43,000
And the universe is infinite in extent.

415
00:22:44,000 --> 00:22:48,000
And there's a uniform finite
density of galaxies throughout the

416
00:22:48,000 --> 00:22:50,000
universe, this density rho.

417
00:22:50,001 --> 00:22:54,000
And each galaxy radiates at
some average power level.

418
00:22:55,000 --> 00:22:58,000
I'm kind of waving your hand
about how evenly distributed

419
00:22:58,000 --> 00:22:59,001
there are, but we really don't care about that.

420
00:23:00,000 --> 00:23:02,001
And then you assume that, oh,
and you look up at the sky at

421
00:23:02,001 --> 00:23:04,000
night, and it's electromagnetic radiation.

422
00:23:04,001 --> 00:23:06,000
It's all free space, path lost to the eye.

423
00:23:06,001 --> 00:23:09,001
And then you do the integral
out to an infinite bound.

424
00:23:11,000 --> 00:23:13,001
Actually, I've screwed up in
the interval a little bit here.

425
00:23:13,001 --> 00:23:17,000
And I was still thinking
2D when I made this slide.

426
00:23:17,001 --> 00:23:19,001
And I don't have the constants right,
but I've got all the important terms.

427
00:23:19,001 --> 00:23:21,000
There's 1 over r squared in here.

428
00:23:22,000 --> 00:23:28,000
And as you're integrating from here out
to infinity distance, you pick up a r

429
00:23:28,000 --> 00:23:30,001
squared dr thing.

430
00:23:31,000 --> 00:23:34,001
And you do the integral to ask,
and the result of the integral is going to tell

431
00:23:34,001 --> 00:23:36,001
you how much power is impinging on your eye.

432
00:23:37,000 --> 00:23:38,000
And the integral diverges.

433
00:23:39,000 --> 00:23:42,000
So it should be an unbounded
infinite amount of power

434
00:23:42,001 --> 00:23:44,000
on your eye when you look
up at the sky at night.

435
00:23:44,001 --> 00:23:45,001
Well,
you're not actually looking at the whole sky.

436
00:23:45,001 --> 00:23:47,000
You're only looking at a piece of it.

437
00:23:47,000 --> 00:23:48,000
But a piece of infinity is still infinity.

438
00:23:50,001 --> 00:23:52,001
And there's something wrong here.

439
00:23:53,000 --> 00:23:55,001
And there's probably something
wrong with one of these

440
00:23:55,001 --> 00:23:58,000
assumptions, or more than one.

441
00:23:59,000 --> 00:24:01,001
And some astronomers,
their standard model for the universe

442
00:24:01,001 --> 00:24:03,000
is that it's actually not infinitely old.

443
00:24:03,000 --> 00:24:05,000
And light travels at a finite speed.

444
00:24:05,001 --> 00:24:09,000
So the universe that we can
see is finite in extent.

445
00:24:10,001 --> 00:24:12,001
And that's why the sky is dark at night, maybe.

446
00:24:13,000 --> 00:24:16,001
They're just theories,
mysteries of the universe.

447
00:24:16,001 --> 00:24:23,000
But anyway, but for packet radio networks,
we live on an infinite flat Earth.

448
00:24:24,000 --> 00:24:27,001
And if we have a uniform finite density
of transmitters throughout, and each

449
00:24:27,001 --> 00:24:33,001
transmitter radiates at a given average power,
alpha is measured.

450
00:24:33,001 --> 00:24:37,000
And I'm just sort of in order to make,
what I've actually got on the integral

451
00:24:37,000 --> 00:24:41,000
here is I've got power as observed
by a receiver at a distance.

452
00:24:41,001 --> 00:24:43,000
That's the power coming out of a thing.

453
00:24:43,001 --> 00:24:44,000
And you do that integral.

454
00:24:45,000 --> 00:24:47,001
And because two-dimensional space
is smaller than three-dimensional

455
00:24:47,001 --> 00:24:50,000
space, the noise shouldn't grow as rapidly.

456
00:24:50,000 --> 00:24:52,000
And hopefully, this integral will not diverge.

457
00:24:54,000 --> 00:24:55,001
Unfortunately, it diverges as well.

458
00:24:56,000 --> 00:24:58,001
You get the integral of 1 over r squared r dr,
which

459
00:24:58,001 --> 00:25:00,000
turns into an integral of 1 over r dr.

460
00:25:01,001 --> 00:25:02,001
And that's log of r.

461
00:25:03,000 --> 00:25:07,000
And you evaluate the
indefinite integral of log r.

462
00:25:07,001 --> 00:25:09,000
And then you plug in the infinite bound.

463
00:25:09,001 --> 00:25:11,000
And log of infinity is still infinity.

464
00:25:12,000 --> 00:25:12,000
What?

465
00:25:14,001 --> 00:25:15,001
Yeah.

466
00:25:17,000 --> 00:25:19,001
So the initial integral diverges as well.

467
00:25:19,001 --> 00:25:22,000
This had me depressed for a couple months.

468
00:25:24,001 --> 00:25:27,001
At this point,
I had decided to do this as a thesis project.

469
00:25:28,000 --> 00:25:32,000
And I was convinced that it all was lost,
because packet radio networks cannot

470
00:25:32,000 --> 00:25:36,001
scale arbitrarily by this result.

471
00:25:40,000 --> 00:25:41,001
But then I learned something.

472
00:25:42,001 --> 00:25:44,001
We do not live on an infinite flat Earth.

473
00:25:46,000 --> 00:25:47,001
But does that really change anything?

474
00:25:49,001 --> 00:25:51,000
We have a finite horizon.

475
00:25:52,000 --> 00:25:52,001
Very good.

476
00:25:52,001 --> 00:25:55,001
So I'm going to do this calculation again.

477
00:25:56,001 --> 00:26:00,001
But I'm going to twist
different knobs this time.

478
00:26:01,000 --> 00:26:03,000
I'm going to say that there's a sixth radius.

479
00:26:03,000 --> 00:26:04,001
This is where my receiver is.

480
00:26:04,001 --> 00:26:06,000
There's some sort of unit circle distance.

481
00:26:06,001 --> 00:26:10,000
And that's where the guy, that's the
distance to the guy I'm trying to listen to.

482
00:26:10,000 --> 00:26:13,001
He's sort of out here at one
unit distance or something.

483
00:26:14,000 --> 00:26:16,001
I'm going to get to that in
just a moment in more detail.

484
00:26:17,000 --> 00:26:22,001
But at this radius, beyond this radius,
they've dropped over the horizon.

485
00:26:22,001 --> 00:26:23,001
And I don't hear them at all.

486
00:26:23,001 --> 00:26:25,001
Or there's enough attenuation
that I can neglect it.

487
00:26:27,001 --> 00:26:28,001
So that's fixed.

488
00:26:29,000 --> 00:26:33,000
And what I'm going to do is I've got
this m is the number of stations.

489
00:26:33,001 --> 00:26:36,000
I didn't use the letter n, because that's
signal to noise and signal to noise.

490
00:26:38,001 --> 00:26:39,001
This radius is fixed.

491
00:26:40,000 --> 00:26:40,001
And m is my knob.

492
00:26:40,001 --> 00:26:41,001
I'm going to turn it up.

493
00:26:41,001 --> 00:26:43,000
I'm going to start out with my two stations.

494
00:26:43,000 --> 00:26:45,001
And I'm going to think about 10 stations, 100

495
00:26:45,001 --> 00:26:47,001
stations, 1,000 stations, and go on.

496
00:26:49,000 --> 00:26:53,000
Now, if we assume this many stations are
sort of uniformly distributed through

497
00:26:53,000 --> 00:26:57,001
the circle, then you take the number of

498
00:26:57,001 --> 00:26:59,000
stations divided by the area of the circle.

499
00:26:59,000 --> 00:27:03,001
And that sort of gives you the density,
number of stations per unit area.

500
00:27:04,001 --> 00:27:09,001
Now, this r sub 0, this is kind of a trick.

501
00:27:11,000 --> 00:27:14,000
What I've done here is I said, well,
how far away is that nearby neighbor?

502
00:27:15,000 --> 00:27:18,000
Well, that kind of depends on
how many stations there are.

503
00:27:19,001 --> 00:27:23,001
And I'm thinking stations that are just
randomly placed, or everybody puts them

504
00:27:23,001 --> 00:27:25,001
on their houses,
and they are over their houses.

505
00:27:25,001 --> 00:27:29,000
But I need some measure of how
far away that nearby neighbor is.

506
00:27:29,001 --> 00:27:34,000
So if you take density, which is number
per unit area, which is numbers per meter

507
00:27:34,000 --> 00:27:38,000
squared or something,
if you invert that and take a square root,

508
00:27:38,000 --> 00:27:39,001
you get something that has units of length.

509
00:27:41,000 --> 00:27:42,000
And I think of that as the character.

510
00:27:42,001 --> 00:27:45,001
Given a density of stations,
that's kind of a characteristic length.

511
00:27:46,001 --> 00:27:52,001
And if you like to think about probability
problems, and you take a circle of

512
00:27:52,001 --> 00:27:57,000
this radius, if you have stations randomly
distributed at this density, and you

513
00:27:57,000 --> 00:28:02,001
think about a circle of this radius r sub 0,
which is the

514
00:28:02,001 --> 00:28:05,000
density raised to the minus 1 half power.

515
00:28:05,000 --> 00:28:06,000
You invert it.

516
00:28:06,001 --> 00:28:08,000
That's 1 over the density square root.

517
00:28:08,001 --> 00:28:10,000
That's 1 over the square root of the density.

518
00:28:11,001 --> 00:28:15,001
If you take a circle of that radius in an
area that you've got randomly stations,

519
00:28:15,001 --> 00:28:17,001
then ask,
what is the expected number of stations

520
00:28:17,001 --> 00:28:19,001
that I'm going to find inside this circle?

521
00:28:19,001 --> 00:28:21,000
That's pi stations.

522
00:28:21,001 --> 00:28:27,000
And if you double that radius, you'd expect
to find 4 pi stations, about a dozen

523
00:28:27,000 --> 00:28:29,000
in a circle.

524
00:28:29,001 --> 00:28:34,001
So I'm going to use this distance here
to represent how far is it to a nearby

525
00:28:34,001 --> 00:28:37,000
neighbor for the purposes of this calculation.

526
00:28:37,001 --> 00:28:42,000
And then there's another factor to
throw in here, which is the duty cycle.

527
00:28:42,001 --> 00:28:45,000
Not all the stations are going
to be transmitting at once.

528
00:28:45,001 --> 00:28:47,001
Maybe only half, or a third, or 10%.

529
00:28:48,001 --> 00:28:50,000
So there's going to be some duty cycle.

530
00:28:50,000 --> 00:28:53,001
So I'm going to carry this duty cycle
through all the analysis as well.

531
00:28:54,001 --> 00:28:56,001
So how much signal is there?

532
00:28:57,000 --> 00:29:03,000
Well, that depends on the transmitter power and

533
00:29:03,000 --> 00:29:04,001
all sorts of constants and stuff.

534
00:29:05,000 --> 00:29:07,000
You can go read about your
antenna and feed line books.

535
00:29:07,001 --> 00:29:10,000
And you've got a feed line attenuation and

536
00:29:10,000 --> 00:29:12,000
perhaps the antenna gain and everything.

537
00:29:12,000 --> 00:29:14,001
But I'm assuming all the stations
are using the same thing.

538
00:29:14,001 --> 00:29:16,001
And so all of the sources of
the interference are the same.

539
00:29:18,000 --> 00:29:25,000
So I'm going to use this constant
alpha to collect all that detail.

540
00:29:25,001 --> 00:29:30,001
And what's important is that the signal
goes as 1 over the distance r squared.

541
00:29:31,001 --> 00:29:34,001
I've got the lambda squared term in here,
but it's all in one band.

542
00:29:35,000 --> 00:29:37,000
What's important is that
this alpha comes in again.

543
00:29:37,001 --> 00:29:39,000
I'm neglecting all the thermal noise.

544
00:29:39,001 --> 00:29:41,001
I'm assuming this is a
self-interference limited system.

545
00:29:43,000 --> 00:29:46,000
So all the interference
comes from similar stations.

546
00:29:46,000 --> 00:29:47,001
So I'm going to have the same alpha down here.

547
00:29:48,000 --> 00:29:51,000
Anyway, you take alpha and 1 over r squared.

548
00:29:51,001 --> 00:29:57,001
Now, if you square this r sub 0,
which is 1 over the square root of the density,

549
00:29:58,000 --> 00:30:01,000
when you square that,
you get this alpha times the density

550
00:30:02,000 --> 00:30:03,001
because this is in the denominator.

551
00:30:04,000 --> 00:30:06,000
So the signal is proportional to the density.

552
00:30:09,000 --> 00:30:11,001
And this rho is a function.

553
00:30:12,000 --> 00:30:17,000
The only thing that's really a variable
in this thing is this capital M, which

554
00:30:17,000 --> 00:30:19,000
on the previous slide.

555
00:30:19,000 --> 00:30:21,001
And this rho is defined as a
function of density as well.

556
00:30:21,001 --> 00:30:23,001
And I'm going to do some
substitution in a moment.

557
00:30:23,001 --> 00:30:26,000
The noise is the same integral again.

558
00:30:26,001 --> 00:30:27,001
It's the Olders paradox integral.

559
00:30:28,000 --> 00:30:31,000
And I'm only going to integrate
out to my fixed radius.

560
00:30:31,000 --> 00:30:34,001
And this r, this rho is also a function of r.

561
00:30:34,001 --> 00:30:38,001
So I've only got one variable to sim,
but I can't.

562
00:30:43,000 --> 00:30:44,000
This r is constant.

563
00:30:44,001 --> 00:30:46,001
This rho is a function of
both the number of stations

564
00:30:46,001 --> 00:30:48,001
and this fixed radius as well.

565
00:30:53,000 --> 00:30:55,000
So I integrate out only to r.

566
00:30:55,000 --> 00:30:56,000
Now, I have another problem.

567
00:30:56,001 --> 00:30:58,000
What do I choose for a lower
bound of the integral?

568
00:30:59,000 --> 00:31:00,001
And I have to hand wave here again.

569
00:31:00,001 --> 00:31:01,000
What's your question?

570
00:31:01,000 --> 00:31:02,000
That's what I was just going to ask.

571
00:31:02,000 --> 00:31:05,000
This is only talking about the
only integrating from the range

572
00:31:05,000 --> 00:31:07,000
of the desired station, the top.

573
00:31:07,000 --> 00:31:08,001
But what about other stations?

574
00:31:08,001 --> 00:31:09,001
That might be closer.

575
00:31:10,000 --> 00:31:10,000
Exactly.

576
00:31:11,000 --> 00:31:13,000
Now, why not just drop this to 0?

577
00:31:13,001 --> 00:31:16,000
This integral also diverges at 0.

578
00:31:16,001 --> 00:31:22,001
It flows up there as well,
because log of 0 is another problem.

579
00:31:23,000 --> 00:31:25,000
So I couldn't go all the way to 0.

580
00:31:25,000 --> 00:31:26,000
I had to pick some number.

581
00:31:26,001 --> 00:31:29,001
The reason I picked r sub 0 is because
it makes the algebra work well.

582
00:31:32,001 --> 00:31:36,001
But I claim it's reasonable, because I've
said the number of stations you expect

583
00:31:36,001 --> 00:31:38,001
to find in here are only pi.

584
00:31:39,000 --> 00:31:42,000
So there's only three other
guys that I'm neglecting here.

585
00:31:42,001 --> 00:31:43,000
Or maybe there's five.

586
00:31:43,000 --> 00:31:45,000
But they're all right next to you.

587
00:31:50,001 --> 00:31:51,001
It doesn't matter.

588
00:31:52,001 --> 00:31:54,000
I'll get the simulation results.

589
00:31:54,000 --> 00:31:56,000
And hopefully by the end of the day,
you won't be worried about this.

590
00:31:57,001 --> 00:31:59,000
Those guys are clearly local.

591
00:31:59,000 --> 00:32:00,001
And maybe we can do some other.

592
00:32:01,000 --> 00:32:02,000
That's a different issue.

593
00:32:03,001 --> 00:32:04,001
Anyway, you do the integral.

594
00:32:05,000 --> 00:32:05,001
And you plug in.

595
00:32:06,000 --> 00:32:07,001
And I'm skipping over a bunch of algebra here.

596
00:32:08,000 --> 00:32:10,000
But you do some simplification.

597
00:32:10,000 --> 00:32:11,001
And you get a term for the noise.

598
00:32:11,001 --> 00:32:13,001
And then you take the ratio of signal to noise.

599
00:32:15,001 --> 00:32:21,001
And everything drops out almost,
leaving only this capital M

600
00:32:21,001 --> 00:32:24,001
and the duty cycle and a few constants.

601
00:32:26,000 --> 00:32:28,000
And this is sort of a remarkable result.

602
00:32:30,000 --> 00:32:34,000
What's sort of stunning to me
when I first saw this was that

603
00:32:34,000 --> 00:32:36,000
the scale length drops out.

604
00:32:36,001 --> 00:32:38,000
I was surprised by that.

605
00:32:38,001 --> 00:32:41,000
After thinking, I realized I shouldn't
have been surprised by that.

606
00:32:41,000 --> 00:32:44,001
Because if you're an interference limited
system and you bring everybody together

607
00:32:44,001 --> 00:32:50,001
by a factor of 2,
then everything goes up by a factor of 4 in

608
00:32:50,001 --> 00:32:53,000
power, both noise and signal.

609
00:32:53,001 --> 00:32:54,001
So it really shouldn't be surprising.

610
00:32:57,000 --> 00:33:00,000
So this is something that's
true at any scale length.

611
00:33:02,001 --> 00:33:05,001
So the signal to noise ratio depends
only on this number of stations.

612
00:33:05,001 --> 00:33:09,000
So it's 1 over the log of how
many stations you've got.

613
00:33:09,001 --> 00:33:12,001
Actually, it's 1 over the duty
cycle times pi times the log

614
00:33:12,001 --> 00:33:14,000
of the number of stations divided by pi.

615
00:33:15,000 --> 00:33:21,001
So the thesis is actually on the web.

616
00:33:22,000 --> 00:33:23,001
There's a sitcom paper on the web.

617
00:33:24,000 --> 00:33:27,001
And the thesis of the PostScript
file is actually put up for FTP.

618
00:33:28,001 --> 00:33:33,000
And somebody at Tapper, before I even
really knew what Tapper was three years

619
00:33:33,000 --> 00:33:39,000
ago, took the PostScript file that I had
made available, converted it to a PDF

620
00:33:39,000 --> 00:33:41,000
file, and put it on the Tapper FTP server.

621
00:33:41,000 --> 00:33:43,000
So there's been a PDF file of my whole thesis.

622
00:33:43,000 --> 00:33:46,000
It's only about 80 pages, single spaced.

623
00:33:48,000 --> 00:33:49,001
It's available on the Tapper website.

624
00:33:50,000 --> 00:33:53,001
And it actually, yeah,
it doesn't skip all the algebra.

625
00:33:54,000 --> 00:33:55,001
So you can actually try and check my algebra.

626
00:34:03,000 --> 00:34:04,000
Go ahead.

627
00:34:04,000 --> 00:34:08,001
The question is, although you derive
it based on circle text, it's now

628
00:34:08,001 --> 00:34:10,001
pretty independent of the one in the case.

629
00:34:11,001 --> 00:34:14,001
A square is almost a circle, I mean,
if you're standing in the middle of it.

630
00:34:14,001 --> 00:34:15,000
[...]

631
00:34:16,000 --> 00:34:16,000
Yeah.

632
00:34:17,000 --> 00:34:20,000
Well, this calculation is in the circle,
because

633
00:34:20,000 --> 00:34:21,001
that's where the pi's and stuff come from.

634
00:34:22,000 --> 00:34:24,000
If it's not a circle, maybe this isn't pi.

635
00:34:24,001 --> 00:34:25,001
Maybe it's 2 or something.

636
00:34:26,000 --> 00:34:30,000
Would it get real messy to try and
introduce some randomness of height, the

637
00:34:30,000 --> 00:34:33,000
concept of height gain,
and the fact of topology, real topology?

638
00:34:33,000 --> 00:34:35,000
Well, I was trying for a worst case.

639
00:34:35,001 --> 00:34:40,000
Anything, any real world propagation
phenomena are going to cut down distant

640
00:34:40,000 --> 00:34:42,000
guys and help you improve.

641
00:34:42,001 --> 00:34:44,001
So this is how bad can it get.

642
00:34:48,000 --> 00:34:51,000
Now, I've got a plot of this for
a bunch of interesting values.

643
00:34:51,001 --> 00:34:54,000
So what I've got on the next slide,
I don't have two projectors, so I

644
00:34:54,000 --> 00:34:55,001
can't show you both at the same time.

645
00:34:55,001 --> 00:34:59,000
But what I'm going to show you is signal
to noise ratio as a function of the

646
00:34:59,000 --> 00:35:02,001
number of stations for a few
different values of duty cycle.

647
00:35:03,001 --> 00:35:04,000
[...]

648
00:35:05,001 --> 00:35:06,000
Let me.

649
00:35:06,000 --> 00:35:12,001
One factor that strikes my mind is
there's obviously a horizontal plane

650
00:35:12,001 --> 00:35:14,000
with which stations can be.

651
00:35:14,001 --> 00:35:17,000
There's also a possibility of a vertical plane.

652
00:35:17,000 --> 00:35:18,000
I neglected that.

653
00:35:18,000 --> 00:35:18,001
That was correct.

654
00:35:20,000 --> 00:35:24,000
But we haven't built
skyscrapers arbitrarily high.

655
00:35:24,000 --> 00:35:25,001
And if you're in a big building like that, you

656
00:35:25,001 --> 00:35:27,000
probably just don't have one
station on the roof anyway.

657
00:35:28,000 --> 00:35:31,001
Actually, I'm trying to get to the suburbs,
not to the downtown.

658
00:35:31,001 --> 00:35:33,001
Do you have a low energy
road on a lot of times?

659
00:35:34,001 --> 00:35:35,000
[...]

660
00:35:35,001 --> 00:35:36,000
Yeah.

661
00:35:37,000 --> 00:35:38,001
Well, because the building attenuates.

662
00:35:39,000 --> 00:35:40,000
I neglected all that.

663
00:35:40,000 --> 00:35:43,001
This is still thinking about stations
in the flat plane of the finite circle.

664
00:35:44,000 --> 00:35:47,000
So here's the plot for signal
to noise ratio on this axis.

665
00:35:48,001 --> 00:35:49,001
Actually, I've got two copies of this plot.

666
00:35:50,000 --> 00:35:51,000
This one I wrote all over.

667
00:35:54,001 --> 00:35:55,001
I'm going to use it.

668
00:35:56,001 --> 00:35:59,000
I've got it labeled in dB,
but then I took a pin.

669
00:35:59,001 --> 00:36:03,001
And this is the 5 dB signal to noise ratio,
or a 3 to 1 ratio.

670
00:36:03,001 --> 00:36:04,001
This is like a 1 to 1 ratio.

671
00:36:05,000 --> 00:36:08,000
This is 0 dB minus 5 minus
10 minus 15 minus 20.

672
00:36:08,001 --> 00:36:11,001
So this would be a 1 part in
100 signal to noise ratio.

673
00:36:12,001 --> 00:36:18,001
And on the x-axis, I have number of stations.
10 to the 2, 10 to the 3, 4,

674
00:36:18,001 --> 00:36:20,000
5, or in other words, this is 1,000.

675
00:36:20,001 --> 00:36:21,000
That's a million.

676
00:36:21,001 --> 00:36:22,001
This is a billion.

677
00:36:23,000 --> 00:36:24,001
And over here would be 10 to the 12.

678
00:36:25,001 --> 00:36:26,000
Stations.

679
00:36:27,001 --> 00:36:31,000
And this curve is 100% duty cycle.

680
00:36:31,001 --> 00:36:33,001
This is a 50% duty cycle, 20%.

681
00:36:33,001 --> 00:36:37,000
And actually, this just moves in dB,
whatever this factor is.

682
00:36:38,000 --> 00:36:40,000
So we can just look at the 100% duty cycle.

683
00:36:41,000 --> 00:36:42,000
All the curves have the same shape.

684
00:36:44,001 --> 00:36:49,000
And so the thing to observe
is that this flattens,

685
00:36:49,000 --> 00:36:50,001
seems to sort of flatten out here.

686
00:36:51,000 --> 00:36:56,000
Which means if you build the station,
if you build the, if you designed

687
00:36:57,001 --> 00:37:03,000
to cope with 1 part in 100 signal
to noise ratios, it would work

688
00:37:03,000 --> 00:37:04,001
here, it would work here.

689
00:37:04,001 --> 00:37:09,000
And then as you got to larger numbers
of stations, it could continue to work.

690
00:37:11,000 --> 00:37:12,001
And this flattens out.

691
00:37:13,000 --> 00:37:19,001
I mean, this 100%
line doesn't not cross the minus 21 dB

692
00:37:19,001 --> 00:37:22,001
line, I believe it was,
until you get to 10 to the 18 stations.

693
00:37:23,000 --> 00:37:23,000
[...]

694
00:37:27,000 --> 00:37:31,000
And this ended my two month
depression about the scaling problem.

695
00:37:31,001 --> 00:37:36,000
This made me think, yes, it can scale
to everybody having one on their roof.

696
00:37:41,000 --> 00:37:47,000
No, they eventually go to
zero signal to noise ratio.

697
00:37:49,001 --> 00:37:52,000
They get there extremely slowly.

698
00:37:52,000 --> 00:37:54,000
Do they go to minus infinity?

699
00:37:55,001 --> 00:37:58,001
Yeah, there's actually, after I finished
the thesis, I did some reading.

700
00:37:59,000 --> 00:38:04,000
And I think in one of,
James Gluck wrote a biography of Richard

701
00:38:04,000 --> 00:38:05,001
Feynman, which I highly recommend.

702
00:38:07,001 --> 00:38:10,001
And there's actually, in the middle of the
book, there's this quote about talking

703
00:38:10,001 --> 00:38:13,001
about when Feynman realized that
a log of infinity is not infinity

704
00:38:13,001 --> 00:38:15,001
in practical sense.

705
00:38:15,001 --> 00:38:18,000
I mean, mathematically, it is.

706
00:38:18,000 --> 00:38:21,001
But any time, because they were solving
some problem in physics, and he just

707
00:38:21,001 --> 00:38:24,000
says, well, what if we just neglect that term,
and then they made progress?

708
00:38:27,000 --> 00:38:27,001
That's kind of neat.

709
00:38:29,000 --> 00:38:32,000
And so logs grow very slowly.

710
00:38:33,000 --> 00:38:38,000
The reason is because we're taking
the log of the number of stations.

711
00:38:38,001 --> 00:38:40,001
And once you logarithm, you have to.

712
00:38:43,001 --> 00:38:44,001
So this is neat.

713
00:38:47,001 --> 00:38:49,001
This, I've got the words on this slide.

714
00:38:49,001 --> 00:38:54,000
The conclusion is you can
hear your nearby neighbor.

715
00:38:55,001 --> 00:38:58,000
And it depends only on the log
and the number of stations.

716
00:38:58,001 --> 00:38:59,001
And it falls very slowly.

717
00:39:01,001 --> 00:39:03,001
And it only gets to about minus 20 dB.

718
00:39:03,001 --> 00:39:05,000
And that's just 100% 3D cycle.

719
00:39:05,001 --> 00:39:08,000
At 50% 3D cycle, things are 3 dB better.

720
00:39:08,001 --> 00:39:10,001
At a 10% 3D cycle, things are 10 dB better.

721
00:39:14,001 --> 00:39:17,001
It's just greater than the
network you can just change.

722
00:39:18,000 --> 00:39:19,000
It depends on the cycle of energy.

723
00:39:21,000 --> 00:39:24,000
Or you can just design to, I mean, yeah.

724
00:39:25,000 --> 00:39:26,000
That's absolutely possible.

725
00:39:26,001 --> 00:39:28,000
There are many possibilities now.

726
00:39:28,000 --> 00:39:31,000
And this made me think there
are many possibilities now.

727
00:39:31,000 --> 00:39:31,001
And I went on.

728
00:39:31,001 --> 00:39:34,001
When you say you have more processes,
does that mean you can have more processes?

729
00:39:35,000 --> 00:39:36,001
[...]

730
00:39:38,001 --> 00:39:39,001
No, you don't actually.

731
00:39:40,000 --> 00:39:44,001
It's a, it turns out, it's a,
the duty cycle issue, it turns out it's a wash.

732
00:39:45,000 --> 00:39:49,000
If you double the duty cycle, if you cut
the duty cycle in half, you get three,

733
00:39:49,001 --> 00:39:51,000
you can only transmit half the time.

734
00:39:51,001 --> 00:39:53,001
But you can get twice as many vistas times.

735
00:39:54,000 --> 00:39:54,001
So it's a wash.

736
00:39:54,001 --> 00:39:56,001
You don't gain anything by
cutting back on duty cycle.

737
00:39:57,000 --> 00:39:59,000
[...]

738
00:39:59,000 --> 00:39:59,001
Yeah.

739
00:40:00,001 --> 00:40:02,000
I didn't come to understand that until later.

740
00:40:03,000 --> 00:40:03,000
But yes.

741
00:40:04,001 --> 00:40:07,001
The question was, well, what's the right
way to place the design of the system?

742
00:40:07,001 --> 00:40:09,000
I didn't plan on mentioning this now.

743
00:40:09,001 --> 00:40:12,000
But something I figured out.

744
00:40:12,000 --> 00:40:14,000
It's like, well,
what duty cycle should you run the system at?

745
00:40:14,000 --> 00:40:20,001
Well, in a noise-limited environment,
the capacity is roughly proportionate.

746
00:40:20,001 --> 00:40:22,001
This is something Matt Aitouz
had on the slide yesterday.

747
00:40:23,000 --> 00:40:24,001
And actually,
I realized I should make that slide.

748
00:40:25,001 --> 00:40:30,000
It says that the bits per second that
you can get through in a noise-limited

749
00:40:30,000 --> 00:40:31,001
system, in other words, with a signal-to-noise

750
00:40:31,001 --> 00:40:33,000
ratio, is noticeably less than one.

751
00:40:34,000 --> 00:40:36,000
It's proportional to the signal-to-noise ratio.

752
00:40:37,000 --> 00:40:42,000
So if you do a fifth, if say you move from
a 20% duty cycle to a 10% duty cycle,

753
00:40:42,001 --> 00:40:45,001
you've improved the
signal-to-noise ratio sort of

754
00:40:45,001 --> 00:40:47,000
throughout the system by a factor of 2.

755
00:40:48,001 --> 00:40:54,000
So you can go twice as fast, but you're
only spending half the time transmitting

756
00:40:54,000 --> 00:40:59,000
the two or the four,
which means that nothing's changed.

757
00:40:59,001 --> 00:41:02,000
So it's a wash as to where you
actually design the system on here.

758
00:41:03,001 --> 00:41:06,000
And in a little while,
I'll be suggesting that you

759
00:41:06,000 --> 00:41:08,000
should spend more than half
the time transmitting.

760
00:41:13,001 --> 00:41:16,001
No, I'm sorry, what?

761
00:41:21,001 --> 00:41:23,000
Yeah, I've lost my design.

762
00:41:24,001 --> 00:41:26,001
You're relaying for people too, so.

763
00:41:28,001 --> 00:41:33,000
So this slide actually covers a lot of ground.

764
00:41:34,001 --> 00:41:36,001
I can't talk about the entire thesis today.

765
00:41:36,001 --> 00:41:40,001
I'm going to have to skip over some details.

766
00:41:40,001 --> 00:41:43,001
But I've still got a lot to talk about.

767
00:41:44,000 --> 00:41:49,001
I think I'm only going to talk about,
well, I need to explain all of this, but

768
00:41:49,001 --> 00:41:52,001
I'm going to hand wave my
way all the way through this

769
00:41:52,001 --> 00:41:54,001
slide to try to give you the whole picture.

770
00:41:55,000 --> 00:41:57,001
And I'm going to come back and
expand on things and try to

771
00:41:57,001 --> 00:41:59,001
illustrate, and hopefully it will all gel.

772
00:42:01,000 --> 00:42:04,001
What I've just showed you is that you can
scale a spread second with sufficient

773
00:42:04,001 --> 00:42:07,000
processing gain is what lets this thing scale.

774
00:42:09,001 --> 00:42:15,001
And you can't, and I understood that it
seems to me that you have to not try to

775
00:42:15,001 --> 00:42:18,000
skip over people,
that you need to route over short hops.

776
00:42:18,001 --> 00:42:21,000
And if somebody shows up between you and
where you were going, you should start

777
00:42:21,000 --> 00:42:24,000
routing through them for the purposes
of letting the system scale.

778
00:42:26,000 --> 00:42:30,000
And I'm going to come back and hopefully
plant some more intuition about

779
00:42:30,000 --> 00:42:31,001
this right towards the end of my talk.

780
00:42:33,000 --> 00:42:36,000
But I was sort of going with
this on intuition at this point.

781
00:42:38,000 --> 00:42:41,001
Every station I expect
will have one transmitter.

782
00:42:44,000 --> 00:42:46,001
When I was thinking through this,
I've got other ideas

783
00:42:46,001 --> 00:42:48,000
for actually how to really do this.

784
00:42:48,001 --> 00:42:53,001
But for now, think of every station
as having one transmitter, one on the

785
00:42:53,001 --> 00:42:59,001
antenna, using that same antenna for receive,
and cannot

786
00:42:59,001 --> 00:43:02,000
hear while its transmitters on.

787
00:43:02,000 --> 00:43:04,000
So when the transmitter comes on,
the station is deaf.

788
00:43:05,000 --> 00:43:09,000
So you can't,
you have to somehow avoid stomping on

789
00:43:09,000 --> 00:43:10,001
yourself on your own reception.

790
00:43:11,001 --> 00:43:13,001
And I'm going to talk in detail about that.

791
00:43:14,001 --> 00:43:16,001
But there's the problem of
what if two people try to

792
00:43:16,001 --> 00:43:18,000
send you a packet at the same time?

793
00:43:18,000 --> 00:43:20,000
Well, with spread spectrum,
you can run multiple of these

794
00:43:20,000 --> 00:43:21,001
spreading channels and capture them both.

795
00:43:24,000 --> 00:43:28,000
And so, for now, I'm going to say
multiple of these spreading channels.

796
00:43:28,000 --> 00:43:30,000
In fact, we'll put in however many we need.

797
00:43:30,001 --> 00:43:34,000
I haven't really,
I haven't come up with the exact

798
00:43:34,000 --> 00:43:35,001
argument for what we really need.

799
00:43:36,000 --> 00:43:38,001
But this is how you solve the
problem of two people sending

800
00:43:38,001 --> 00:43:40,000
you packets at the same time.

801
00:43:40,001 --> 00:43:46,001
A lot of literature on packet rate,
like Macaw, MACA, and then all of the follow

802
00:43:46,001 --> 00:43:51,000
on, the MACAW and the FANA
stuff that's all been done just

803
00:43:51,000 --> 00:43:52,001
on the last eight or so years.

804
00:43:53,000 --> 00:43:56,001
Starting with Phil Carnes paper,
which was a DCC paper in 1990, I think.

805
00:43:56,001 --> 00:44:01,000
I'm not really thinking about Indian

806
00:44:01,000 --> 00:44:03,000
protocols,
I'm just trying to get communication working.

807
00:44:03,000 --> 00:44:08,000
But the, Is there some of these issues of,
you know, propagation delay with

808
00:44:08,000 --> 00:44:11,000
driving protocols and propagation delay?

809
00:44:11,000 --> 00:44:14,000
Well,
propagation is five microseconds per mile.

810
00:44:15,000 --> 00:44:16,000
And I sort of neglect that.

811
00:44:17,000 --> 00:44:19,001
You have also got issue of
protocol acknowledgment delays.

812
00:44:20,000 --> 00:44:23,000
Well,
I'm thinking I want to do end to end protocols.

813
00:44:24,000 --> 00:44:27,001
So I'm going to, I'd like to be able to
engineer the system so that the per hop

814
00:44:27,001 --> 00:44:32,000
reliability is high enough so that we can
just build the reliability on the end

815
00:44:32,000 --> 00:44:35,000
end things like the rest of the internet works.

816
00:44:36,001 --> 00:44:42,000
So I'd like to, when a packet is
transmitted by a station, I'd like it to be

817
00:44:42,000 --> 00:44:44,001
received by that receiver almost certainly.

818
00:44:45,000 --> 00:44:48,000
So I don't need to do any hop by
hop acknowledgment, any feedback.

819
00:44:48,001 --> 00:44:50,000
And so you can just go on.

820
00:44:50,001 --> 00:44:53,001
And then do that,
you have to get rid of any packet loss

821
00:44:53,001 --> 00:44:55,000
due to what people think of as collisions.

822
00:44:56,001 --> 00:45:00,001
Collisions, I think are a misleading term
because, you know, it's all signal to

823
00:45:00,001 --> 00:45:04,001
noise ratios and all depends on, you know,
whether you had enough to decode it.

824
00:45:05,000 --> 00:45:10,000
And if you're in the middle of a sea of
a million stations, then the real issue

825
00:45:10,000 --> 00:45:14,001
is if you have enough processing gain and,
you know, whether the signal noise

826
00:45:14,001 --> 00:45:18,000
ratio or decoding, what was the
probability of decoding that packet?

827
00:45:18,000 --> 00:45:20,001
But I'd like to engineer that to be reliable.

828
00:45:21,000 --> 00:45:23,000
And I don't really think about the real

829
00:45:23,000 --> 00:45:24,001
problem that I'm trying to solve as collisions.

830
00:45:24,001 --> 00:45:26,000
I didn't come at this from
a collisions problem.

831
00:45:26,000 --> 00:45:29,000
But if we want to think about collisions
for a second, there's three cases.

832
00:45:29,001 --> 00:45:32,000
I don't have the slide for this,
but a little picture in the thesis.

833
00:45:33,000 --> 00:45:36,001
There's a, I don't remember what number,
what order I put

834
00:45:36,001 --> 00:45:38,001
them in the thesis,
but there's a couple possibilities.

835
00:45:38,001 --> 00:45:44,000
There's like, I'm trying to receive a
packet from a guy over here, and then

836
00:45:44,000 --> 00:45:46,001
there's a guy over here trying to
send a packet to somebody else.

837
00:45:46,001 --> 00:45:50,000
And so I get a signal from him,
and I'm getting interference from this

838
00:45:50,000 --> 00:45:52,001
transmission for going over there.

839
00:45:53,000 --> 00:45:54,000
That's one type of collision.

840
00:45:54,001 --> 00:45:56,001
And with spread spectrum,
I mean, I've already like

841
00:45:56,001 --> 00:45:58,001
included like all the other transmissions.

842
00:45:59,000 --> 00:46:03,000
And so if this is a sort of a type one
collision, I'm not really worried about

843
00:46:03,000 --> 00:46:05,000
that because I'm sort of solving
that with spread spectrum.

844
00:46:05,001 --> 00:46:09,000
The second case would be two people
sending packets at the same time to me.

845
00:46:09,000 --> 00:46:12,001
And that's what this multiple
de-spreading channels issue is about.

846
00:46:12,001 --> 00:46:16,001
And another type of collision is somebody's
trying to send a packet to me, and at

847
00:46:16,001 --> 00:46:18,001
the same time I'm transmitting
a packet to somebody else.

848
00:46:19,000 --> 00:46:23,001
And that is still a problem because I'm
completely deaf when my transmitter's on

849
00:46:23,001 --> 00:46:26,001
because the most performance you
can get out of a duplexer for

850
00:46:26,001 --> 00:46:28,000
an antenna is about 20 dB isolation.

851
00:46:28,001 --> 00:46:32,000
And the difference between a transmitted
signal level up the antenna and what

852
00:46:32,000 --> 00:46:34,001
comes down at the other end
is on the order of 100 dB.

853
00:46:37,000 --> 00:46:41,000
So you can only get about 20 dB of
isolation on the same frequency.

854
00:46:42,001 --> 00:46:43,000
Yeah.

855
00:46:44,000 --> 00:46:48,000
Well, the old cards are getting senior,
you know, 90, but that's the likelihood.

856
00:46:48,001 --> 00:46:50,000
Well, in separate antennas.

857
00:46:50,001 --> 00:46:53,000
They're not using the same feed line for that.

858
00:46:53,000 --> 00:46:55,000
Well, you can't get any separation.

859
00:46:56,000 --> 00:46:59,001
You can get about 20 dB on the same feed line,
but only 20 and you need about

860
00:46:59,001 --> 00:47:01,000
100, so you're about 80 dB short.

861
00:47:01,000 --> 00:47:04,000
There are some short order systems for dB,
but they're very narrow banded.

862
00:47:04,001 --> 00:47:05,000
And probably extensive.

863
00:47:05,000 --> 00:47:07,001
So, wait a minute.

864
00:47:07,001 --> 00:47:14,000
If you go back to that 100% duty cycle,
is everybody's transmitting 100%?

865
00:47:14,001 --> 00:47:16,000
[...] do you detect only everybody's deaf?

866
00:47:17,000 --> 00:47:17,000
That's right.

867
00:47:18,000 --> 00:47:19,001
So people are going to...

868
00:47:19,001 --> 00:47:19,001
What?

869
00:47:20,000 --> 00:47:20,001
No, it has to be less.

870
00:47:21,000 --> 00:47:21,001
That's right.

871
00:47:22,001 --> 00:47:23,000
That's right.

872
00:47:25,000 --> 00:47:26,000
But maybe 50%.

873
00:47:26,001 --> 00:47:29,001
That's only 3 dB difference
in the whole noise level.

874
00:47:30,001 --> 00:47:32,000
I don't think I said anything contrary to that.

875
00:47:33,000 --> 00:47:38,001
Are you assuming that the noise calculation,
which hypothesis you gave

876
00:47:40,000 --> 00:47:44,000
against all other stations,
all the stations within radius r?

877
00:47:48,001 --> 00:47:52,000
Well, to be truly orthogonal, it all have to be

878
00:47:52,000 --> 00:47:54,000
synchronized from the point
of view of the receiver.

879
00:47:55,000 --> 00:47:55,001
I mean...

880
00:47:55,001 --> 00:47:57,000
Within 3 [...]

881
00:47:57,000 --> 00:47:59,001
Well, perhaps there are...

882
00:47:59,001 --> 00:48:02,000
I mean,
do you consider random codes to be orthogonal?

883
00:48:05,001 --> 00:48:10,001
What kind of performance do you get
if everybody uses codes that...

884
00:48:10,001 --> 00:48:11,000
The code...

885
00:48:11,001 --> 00:48:14,001
There's a lot of detail
that we could go into here.

886
00:48:17,000 --> 00:48:21,000
What do you mean by orthogonal?

887
00:48:22,000 --> 00:48:25,001
Meaning that the correlation...

888
00:48:25,001 --> 00:48:32,000
[...] correlation... ...
the system in there... [...] amount

889
00:48:32,000 --> 00:48:36,000
located in the correlation range...
...and the station [...]

890
00:48:36,001 --> 00:48:37,000
[...]

891
00:48:37,001 --> 00:48:43,000
In which case, the number of orthogonal
codes you need... ...is equal to the

892
00:48:43,000 --> 00:48:44,001
number of stations...

893
00:48:46,001 --> 00:48:47,000
Yeah.

894
00:48:47,001 --> 00:48:48,001
There's a sort of a...

895
00:48:48,001 --> 00:48:51,000
There's a standard way that
spread spectrum systems

896
00:48:51,000 --> 00:48:52,001
are built where everybody uses...

897
00:48:56,001 --> 00:48:59,001
They repeat the same chip
pattern for every bit.

898
00:49:00,000 --> 00:49:03,001
Basically, the spreading code repetition
rate is the same as the bit time.

899
00:49:04,000 --> 00:49:05,000
And you don't need to do that.

900
00:49:05,001 --> 00:49:06,000
So you can...

901
00:49:06,001 --> 00:49:09,001
You can have spreading codes
that repeat over longer cycles.

902
00:49:10,001 --> 00:49:14,000
And when you go to that, then you
actually have a larger number of codes.

903
00:49:15,000 --> 00:49:18,000
And now you have the problem of like,
well, did I get this bit right?

904
00:49:18,000 --> 00:49:19,001
And it's like, well, there...

905
00:49:19,001 --> 00:49:21,000
That's the probabilistic thing.

906
00:49:21,001 --> 00:49:23,001
And you're probably going to have to
wrap error correcting codes around that.

907
00:49:23,001 --> 00:49:28,001
And I appeal to the, you know,
if you treat people with noise in the band...

908
00:49:29,000 --> 00:49:31,000
...and, you know, Shannon... ...Saul Shannon's

909
00:49:31,000 --> 00:49:32,001
equation,
then there ought to be some way to do it.

910
00:49:33,000 --> 00:49:34,000
I haven't...

911
00:49:34,000 --> 00:49:37,001
The literature talks about receiver-directed
codes, transmitter-directed codes.

912
00:49:38,001 --> 00:49:41,000
I wave my hands and I say,
everybody's using random codes.

913
00:49:42,000 --> 00:49:46,001
And when you're talking 20 dB of processing
gain, you've got on the order of...

914
00:49:47,001 --> 00:49:52,001
You know, if you did it that way,
you'd have on the order of 100 chips per bit.

915
00:49:53,001 --> 00:49:58,000
And even if you want to constrain so
that the offset in any bit is at most

916
00:49:58,000 --> 00:49:59,001
40, 60, the number of...

917
00:50:00,001 --> 00:50:01,001
There's a lot of space there.

918
00:50:02,000 --> 00:50:05,000
So I haven't really worried
much about the number of codes.

919
00:50:08,001 --> 00:50:15,000
I want to solve it all,
but I had to turn in the thesis at some point.

920
00:50:19,000 --> 00:50:21,001
So multiple of these spreading
channels to solve a problem of

921
00:50:21,001 --> 00:50:23,000
people sending in multiple packets within time.

922
00:50:23,000 --> 00:50:25,000
And then the remaining collisions are

923
00:50:25,000 --> 00:50:27,000
those which are caused by your own transmitter.

924
00:50:27,000 --> 00:50:32,000
And I'm going to show you a scheduling
method right now after the slide.

925
00:50:32,001 --> 00:50:33,000
How do I do that?

926
00:50:33,001 --> 00:50:38,001
Maybe six, maybe eight.

927
00:50:40,000 --> 00:50:40,001
And GPS...

928
00:50:40,001 --> 00:50:45,000
I mean, once you actually go to VLSI and
you're not actually trying to do it all

929
00:50:45,000 --> 00:50:48,000
digitally... ...but you actually got
analog circuits with buckets of integrate

930
00:50:48,000 --> 00:50:50,000
and dump and everything... ...it's not

931
00:50:50,000 --> 00:50:51,001
hard to replicate these spreading channels.

932
00:50:53,001 --> 00:50:55,001
You can share all the receiver front end.

933
00:50:59,000 --> 00:51:00,001
Now, power control...

934
00:51:02,000 --> 00:51:05,000
The method I wound up using, sort of...

935
00:51:05,000 --> 00:51:07,000
It was an ad hoc choice,
but it turned out to work really well.

936
00:51:07,001 --> 00:51:11,001
And the simulation I sort of believe in
is to choose a power level to deliver to

937
00:51:11,001 --> 00:51:14,000
the receiver... ...and then adjust
the transmitter level to do that.

938
00:51:14,001 --> 00:51:17,001
And the stations will somehow rendezvous
and learn where each other are.

939
00:51:17,001 --> 00:51:20,001
They can measure the propagation
characteristics between

940
00:51:20,001 --> 00:51:22,000
them and then adjust appropriately.

941
00:51:22,001 --> 00:51:28,000
And so actually, in my simulations,
I controlled power to deliver minus 120

942
00:51:28,000 --> 00:51:32,000
dBw... ...and to the receiver at the far end.

943
00:51:33,000 --> 00:51:33,001
Assuming isotropic...

944
00:51:33,001 --> 00:51:36,000
I mean, I also used isotropic
antennas to handle anything.

945
00:51:36,001 --> 00:51:40,000
You know, it really doesn't matter because...

946
00:51:40,000 --> 00:51:41,000
Well, I'll show you something in a moment.

947
00:51:41,001 --> 00:51:43,001
And then how to do routing...

948
00:51:44,001 --> 00:51:49,000
Actually, Phil Carnes' paper in 1991
mentions minimum energy routing... ...and

949
00:51:49,000 --> 00:51:51,000
credits it to Dave Mills,
who's on the faculty at Maryland.

950
00:51:52,001 --> 00:51:55,001
And Dave Mills got this intuition
from thinking about a whole bunch of

951
00:51:55,001 --> 00:51:58,001
satellites... ...that are trying
to form a network and communicate.

952
00:51:59,001 --> 00:52:01,001
And that minimum energy routing
is the right thing to do.

953
00:52:02,001 --> 00:52:05,000
And so that's what I started with.

954
00:52:06,000 --> 00:52:10,000
When I started sort of trying to
design this system through... ...I

955
00:52:10,000 --> 00:52:11,001
thought, well,
this is what I'm going to use for

956
00:52:11,001 --> 00:52:13,001
now... ...
but there's probably something more optimum.

957
00:52:14,000 --> 00:52:16,001
And when I get to the end, I'll try to find...
...figure out what's better.

958
00:52:17,000 --> 00:52:19,000
And I'll get to that towards
the end of the talk.

959
00:52:19,001 --> 00:52:23,001
But this is sort of a whole picture
of what I'm expecting to do.

960
00:52:24,000 --> 00:52:27,001
Now, I'm going to talk about... ...
avoid remaining collisions by scheduling

961
00:52:27,001 --> 00:52:29,001
and how that works.

962
00:52:30,000 --> 00:52:31,001
And this is how I do it.

963
00:52:34,000 --> 00:52:38,000
Before I tell you how I do it, though...
...you can open, like, the Burt Segges

964
00:52:38,000 --> 00:52:42,000
and Gallagher textbook... ...and turn to
the chapter or section on packet radio

965
00:52:42,000 --> 00:52:44,000
networks... ...
and it explains what a packet radio network is.

966
00:52:44,000 --> 00:52:47,000
I think it only goes through about three
or four pages on packet radio networks.

967
00:52:47,001 --> 00:52:49,000
But it talks about the problem of collisions.

968
00:52:49,001 --> 00:52:53,000
And that the way you solve this is you
figure out which nodes can hear each

969
00:52:53,000 --> 00:52:57,001
other... ...and guarantee that, you know,
at any receiver only one transmitter

970
00:52:57,001 --> 00:53:00,000
is transmitting.

971
00:53:00,001 --> 00:53:03,000
You know, at the intended receiver,
that receiver can only hear... ...the

972
00:53:03,000 --> 00:53:05,000
one transmitter that's sending to it.

973
00:53:05,000 --> 00:53:08,001
And all of...you know, you don't have
any other adjacent nodes in the network.

974
00:53:09,000 --> 00:53:10,001
And that the way you solve that problem

975
00:53:10,001 --> 00:53:12,000
is you have synchronized clocks everywhere.

976
00:53:12,000 --> 00:53:15,000
You figure out what the topology is...
...and you come up with a master schedule

977
00:53:15,000 --> 00:53:17,001
that just repeats every once in a while...
...that everybody sticks to

978
00:53:17,001 --> 00:53:19,000
that says who can send to who when.

979
00:53:23,000 --> 00:53:25,001
And...I mean, I have a few problems with that.

980
00:53:25,001 --> 00:53:29,000
One is the idea that propagation is...
...you can view the whole thing as a

981
00:53:29,000 --> 00:53:30,001
node and edges sort of situation.

982
00:53:30,001 --> 00:53:32,000
I mean, it's more complicated than that.

983
00:53:33,001 --> 00:53:36,001
But the real problem is,
is there are a million stations out

984
00:53:36,001 --> 00:53:38,000
there... ...do you ever think we
can get them all synchronized?

985
00:53:38,001 --> 00:53:41,001
And even if we could get them all synchronized.
.. ...do you think we could do the

986
00:53:41,001 --> 00:53:44,000
computation to figure out
what the master schedule

987
00:53:44,000 --> 00:53:45,001
is... ...and get it out to everybody?

988
00:53:46,000 --> 00:53:50,001
What I want is I want... ...
I'd rather not have the system

989
00:53:50,001 --> 00:53:52,001
depend on GPS continuing to work.

990
00:53:54,000 --> 00:53:55,001
So...that's the GPS problem.

991
00:53:56,000 --> 00:53:59,000
So even if we could get them synchronized...
...I still don't have much

992
00:53:59,000 --> 00:54:01,000
confidence in being able to get a
master schedule out to everybody.

993
00:54:01,000 --> 00:54:05,001
And these master schedules have the
property that... ...the fact that

994
00:54:05,001 --> 00:54:10,001
connectivity has changed 30 hops away from you.
.. ...affects everybody and then

995
00:54:10,001 --> 00:54:13,001
all the constraints keep propagating...
...and you wind up having to change the

996
00:54:13,001 --> 00:54:15,000
schedule globally for every little change.

997
00:54:15,001 --> 00:54:19,000
And people have tried to tackle that,
but I wanted something simpler.

998
00:54:19,001 --> 00:54:21,000
How do people do that type of schedule?

999
00:54:22,001 --> 00:54:24,000
I'm busy now, I'll come back in an hour.

1000
00:54:24,001 --> 00:54:25,001
That's kind of what I was thinking about.

1001
00:54:26,001 --> 00:54:27,000
Yeah.

1002
00:54:28,001 --> 00:54:31,000
Well, I'll tell you what I came up with.

1003
00:54:33,000 --> 00:54:35,000
It wasn't the slide I expected.

1004
00:54:37,001 --> 00:54:39,000
I thought I had another slide here.

1005
00:54:41,001 --> 00:54:42,001
Maybe...no, this is the one.

1006
00:54:46,000 --> 00:54:52,001
Every station generates a pseudo-random
schedule of... ...when it will transmit

1007
00:54:52,001 --> 00:54:57,000
and when it will receive...
...off into the future.

1008
00:55:01,001 --> 00:55:03,001
No, no. I now...I lay down.

1009
00:55:04,001 --> 00:55:08,001
At this time I will be transmitting,
at this time I will be listening, at this

1010
00:55:08,001 --> 00:55:10,000
time I will be transmitting,
at this time I will be listening.

1011
00:55:11,000 --> 00:55:12,000
But you make it pseudo-random.

1012
00:55:12,001 --> 00:55:16,000
The reason you make it pseudo-random is
if I just said, okay, during odd seconds

1013
00:55:16,000 --> 00:55:20,001
I will transmit... ...and during even
seconds I will receive... ...then you have

1014
00:55:20,001 --> 00:55:24,001
the problem that somebody next to me can
do that... ...then you have the problem

1015
00:55:24,001 --> 00:55:26,000
that you have eight type A stations and type

1016
00:55:26,000 --> 00:55:28,000
B stations and not everybody
can talk to everybody.

1017
00:55:28,000 --> 00:55:32,000
If they're making them pseudo-random,
then everybody...then there will

1018
00:55:32,000 --> 00:55:33,001
be a chance in the future.

1019
00:55:34,000 --> 00:55:35,001
You know,
for any pair of stations there will be

1020
00:55:35,001 --> 00:55:37,000
times when they can each talk to each other.

1021
00:55:45,000 --> 00:55:46,001
Well, there will be times.

1022
00:55:47,000 --> 00:55:52,000
So, you do a pseudo-random
schedule off into the future.

1023
00:55:52,001 --> 00:55:55,000
I do a pseudo-random schedule
off into the future.

1024
00:55:55,001 --> 00:55:57,000
There will be times when your schedule and my

1025
00:55:57,000 --> 00:55:58,001
schedule says you transmit, I transmit.

1026
00:55:59,000 --> 00:56:01,001
And there will be times in the future
where it says you receive, I receive.

1027
00:56:01,001 --> 00:56:03,001
But there will be other times when
it says you transmit, I receive.

1028
00:56:03,001 --> 00:56:05,000
And there will be other times
when it says vice versa.

1029
00:56:06,000 --> 00:56:09,001
You accept that there are going to
be conflicts in those pseudo-rands.

1030
00:56:09,001 --> 00:56:12,000
Right. There will be times
when I cannot transmit to you.

1031
00:56:12,001 --> 00:56:15,000
There will be times when neither
of us can talk to each other.

1032
00:56:15,000 --> 00:56:17,000
All these cases know that neither is scheduled.

1033
00:56:18,000 --> 00:56:19,000
Right. And so what you do...

1034
00:56:19,000 --> 00:56:23,001
And all you need to do is get
this schedule to the neighbor

1035
00:56:23,001 --> 00:56:25,000
that you want to communicate with directly.

1036
00:56:26,000 --> 00:56:30,000
And by doing it this way, that has no
impact on, you know, if I hand you a

1037
00:56:30,000 --> 00:56:32,000
schedule and you hand me my schedule.

1038
00:56:32,001 --> 00:56:35,001
So now we've had an interaction
and now we can talk to each other.

1039
00:56:36,000 --> 00:56:41,000
And what we've done to arrange to talk
to each other has no impact on your

1040
00:56:41,000 --> 00:56:43,000
negotiations with somebody else.

1041
00:56:43,000 --> 00:56:44,001
In fact, there's no negotiations. You just

1042
00:56:44,001 --> 00:56:46,000
tell your neighbor what your schedule is.

1043
00:56:46,000 --> 00:56:47,001
Oh, you have to transmit.

1044
00:56:49,000 --> 00:56:49,001
[...]

1045
00:56:50,000 --> 00:56:51,001
You only have to transmit three.

1046
00:56:51,001 --> 00:56:55,000
Well, yes. I'll get to that. Two more slides.

1047
00:56:55,001 --> 00:56:58,001
So I have a picture. Now, I don't...

1048
00:56:58,001 --> 00:57:02,001
I'm not even assuming that we
have slot times synchronized.

1049
00:57:03,001 --> 00:57:08,000
You may think your watch may have the
seconds starting... 300 milliseconds after

1050
00:57:08,000 --> 00:57:10,000
my watch. That's okay.

1051
00:57:11,000 --> 00:57:14,000
As long as when I communicate the
schedule to you, you understand on

1052
00:57:14,000 --> 00:57:16,000
what time base I'm getting that on.

1053
00:57:16,000 --> 00:57:18,001
And then you can reckon it
by your time base and figure

1054
00:57:18,001 --> 00:57:20,000
out where the opportunities are.

1055
00:57:20,001 --> 00:57:24,000
So here's a bunch of schedules
and actually it's a little small,

1056
00:57:24,001 --> 00:57:27,000
but they're not synchronized.

1057
00:57:28,001 --> 00:57:31,001
But if you want to send a packet,
you can find a spot.

1058
00:57:33,000 --> 00:57:39,000
You might have to wait a while, you know,
but you'll be able to find a spot.

1059
00:57:39,001 --> 00:57:42,001
What are the ramifications on real time?

1060
00:57:42,001 --> 00:57:46,000
Delay.
At the very end, I'll plug in some numbers.

1061
00:57:47,001 --> 00:57:51,000
The thesis has some formulas you can
like plug in your own parameters for a

1062
00:57:51,000 --> 00:57:52,001
system and figure out what the delays are.

1063
00:57:53,000 --> 00:57:55,001
I'll plug in... I'll show you some
results of plugged in numbers at the end.

1064
00:57:57,001 --> 00:57:59,000
So that's the pseudo random scheduling method.

1065
00:57:59,001 --> 00:58:03,001
Everybody generates a random
schedule and just hands that.

1066
00:58:03,001 --> 00:58:08,000
Now, the implementation trick for that
is all stations use the same schedule.

1067
00:58:10,000 --> 00:58:13,001
But everybody sets their clock to
a random value when they turn on.

1068
00:58:14,000 --> 00:58:15,000
Now, you may be concerned about...

1069
00:58:16,000 --> 00:58:20,000
And then all you have to do is hand
what your clock is to your neighbor.

1070
00:58:20,000 --> 00:58:23,001
And now you know what the offset is to
what time your neighbor thinks it is.

1071
00:58:25,001 --> 00:58:28,001
And you remember that and then you
can tell what the schedule is.

1072
00:58:28,001 --> 00:58:31,000
And maybe your clocks run at
slightly different rates.

1073
00:58:31,000 --> 00:58:33,000
Well, that's something you can model as well.

1074
00:58:33,001 --> 00:58:36,001
If my clock's running faster and I've
swapped time with you a few times,

1075
00:58:36,001 --> 00:58:38,000
then you can actually set a line to that.

1076
00:58:38,001 --> 00:58:40,001
And maybe your crystal's aging
at a different rate than mine.

1077
00:58:41,000 --> 00:58:43,000
Well,
you subtract those and you get a quadratic.

1078
00:58:43,000 --> 00:58:46,000
And you put a quadratic to that
and then you can get amazing

1079
00:58:46,000 --> 00:58:48,000
stability that lasts over days maybe.

1080
00:58:48,001 --> 00:58:49,001
But you don't even need to do that.

1081
00:58:49,001 --> 00:58:51,001
I mean, or you could just track me.

1082
00:58:52,000 --> 00:58:54,000
You know, when we re-run these over as we

1083
00:58:54,000 --> 00:58:56,000
communicate,
you can sort of figure out where I'm going.

1084
00:58:58,001 --> 00:59:04,000
Now, you may be concerned that, well,
what if we set our clock to the same value?

1085
00:59:05,000 --> 00:59:09,000
Well, you can just make enough high
order bits in the clock to push the

1086
00:59:09,000 --> 00:59:11,001
probability of that happening arbitrarily low.

1087
00:59:12,001 --> 00:59:18,000
If we exchange 64-bit clocks, the chances
that we actually, you know, got close

1088
00:59:18,000 --> 00:59:23,000
enough to worry about overlapping exactly,
you know, synchronizing our

1089
00:59:23,000 --> 00:59:24,001
schedules anytime soon is very low.

1090
00:59:25,000 --> 00:59:28,001
And all you really care,
you don't care about if somebody across town

1091
00:59:28,001 --> 00:59:30,000
is exactly on your same schedule.

1092
00:59:30,000 --> 00:59:30,001
That doesn't matter.

1093
00:59:30,001 --> 00:59:33,001
All that matters is that about the six
or so people that you're in direct

1094
00:59:33,001 --> 00:59:37,000
communication with didn't happen
to land on the same schedule.

1095
00:59:37,001 --> 00:59:38,001
Well, here's the notion.

1096
00:59:39,001 --> 00:59:44,001
When you initially plug in, maybe you
could select clock values of all neighbors

1097
00:59:44,001 --> 00:59:47,001
you hear and then figure
out an appropriate clock

1098
00:59:47,001 --> 00:59:49,000
value for yourself based on that.

1099
00:59:49,001 --> 00:59:51,000
Yeah, but then you have the
problem somebody else shows up.

1100
00:59:51,001 --> 00:59:53,001
It's easy just to add bits
to the left of the clock.

1101
00:59:54,001 --> 00:59:58,000
Every new bit you get is only one more
bit to hand to your neighbor, but it

1102
00:59:58,000 --> 00:59:59,001
produces the probability by two.

1103
01:00:00,000 --> 01:00:04,000
If you [...] in for them,
there are a backhand direction on that one.

1104
01:00:05,000 --> 01:00:08,000
So the reality is everybody
wouldn't have a comparison.

1105
01:00:09,000 --> 01:00:11,001
You can, it's a birthday paradox sort of thing.

1106
01:00:12,000 --> 01:00:15,000
You can actually put enough bits at the
left so that no two in the world never

1107
01:00:15,000 --> 01:00:17,001
get close enough to each other,
and you could actually make a bet on that.

1108
01:00:19,000 --> 01:00:22,000
I mean, you can't guarantee, because
mathematically it's still possible, that

1109
01:00:22,000 --> 01:00:24,000
you can actually make the bet.

1110
01:00:24,000 --> 01:00:26,000
It's like the probabilistic test to test

1111
01:00:26,000 --> 01:00:28,000
whether a number is prime
when they do cryptography.

1112
01:00:28,000 --> 01:00:33,001
They don't really ever quite prove that
you're using a prime number to build your

1113
01:00:33,001 --> 01:00:35,001
key for RSA encryption.

1114
01:00:36,000 --> 01:00:41,000
But you can be so sure that, you know,
you can be as sure as you like.

1115
01:00:41,000 --> 01:00:44,001
Just by every time you run the test again,
the test runs quickly.

1116
01:00:45,000 --> 01:00:47,001
So that's the implementation trick.

1117
01:00:47,001 --> 01:00:50,000
Now, there's a question that is, well,
okay, when you generate these pseudo

1118
01:00:50,000 --> 01:00:53,000
-random schedules, actually I've said
everybody's going to use the same schedule.

1119
01:00:53,000 --> 01:00:57,001
So when we figure out what that one schedule
is, what fraction of the slot should

1120
01:00:57,001 --> 01:01:01,001
be received, and what fraction of
the slot should be transmitted?

1121
01:01:04,000 --> 01:01:08,001
And, well,
I think of this as the received duty cycle.

1122
01:01:08,001 --> 01:01:10,000
What fraction of the slots are received slots?

1123
01:01:10,001 --> 01:01:11,001
I call that P.

1124
01:01:13,000 --> 01:01:18,000
And the fraction of time that I can send
to a particular neighbor, if it's a good

1125
01:01:18,000 --> 01:01:21,000
hash function that generates the
pseudo-random schedule, then you can think of

1126
01:01:21,000 --> 01:01:22,001
them as independent events.

1127
01:01:23,000 --> 01:01:25,001
And so you can do this sort of
simple probabilistic analysis.

1128
01:01:26,000 --> 01:01:31,001
So the fraction of time you can send to
a particular neighbor is the probability

1129
01:01:31,001 --> 01:01:36,000
that your neighbor is listening,
which is P, the chance that it's a received

1130
01:01:36,000 --> 01:01:37,001
duty cycle, at any given moment.

1131
01:01:39,001 --> 01:01:45,001
And the probability that you're in a
transmit slot, which is 1 minus P.

1132
01:01:47,000 --> 01:01:48,000
You multiply those two together.

1133
01:01:49,000 --> 01:01:54,001
And then there's actually a formula,
skipping over some algebra.

1134
01:01:55,000 --> 01:01:58,001
So you can say, well,
what fraction of the time is it

1135
01:01:58,001 --> 01:02:00,001
possible to send to a neighbor?

1136
01:02:00,001 --> 01:02:05,000
Assume you have traffic for all your neighbors,
and you want to know, well, how

1137
01:02:05,000 --> 01:02:08,001
much can I keep my transmitter going?

1138
01:02:10,000 --> 01:02:11,001
And that's the formula here.

1139
01:02:11,001 --> 01:02:14,001
And it's basically, the fraction of
time you can spend transmitting to your

1140
01:02:14,001 --> 01:02:20,001
neighbors is the fraction of time that
your schedule says you can transmit, minus

1141
01:02:20,001 --> 01:02:27,000
the probability that you're a transmitter,
that your schedule says transmit,

1142
01:02:27,000 --> 01:02:29,001
and everybody else's schedule says transmit.

1143
01:02:30,000 --> 01:02:32,000
And that's why it's K plus 1 for K neighbors.

1144
01:02:34,000 --> 01:02:40,001
And so what I've got on the next chart is
a plot of this for various values of K.

1145
01:02:43,000 --> 01:02:49,001
This is the fraction of time you can
spend sending to a neighbor for received

1146
01:02:49,001 --> 01:02:52,000
duty cycles from 0 all the way to 1.

1147
01:02:54,001 --> 01:02:58,001
And if you only have one neighbor,
then the best value to pick is

1148
01:02:58,001 --> 01:03:00,001
50% received duty cycle.

1149
01:03:01,000 --> 01:03:04,001
And that means the fraction of time you
can send to him is 1 fourth, and the

1150
01:03:04,001 --> 01:03:07,000
fraction of time that he can
send to you is also 1 fourth.

1151
01:03:08,001 --> 01:03:11,001
If you have two neighbors,
you get a curve that looks like this.

1152
01:03:12,000 --> 01:03:15,001
This is the fraction of time that you
can send to, it's the or of the two.

1153
01:03:15,001 --> 01:03:18,000
When can I send to this neighbor,
and when can I send to that neighbor?

1154
01:03:18,000 --> 01:03:19,001
When can I send to either of them?

1155
01:03:20,001 --> 01:03:22,001
You know, is at least one or both listening.

1156
01:03:23,001 --> 01:03:24,000
And that's this curve.

1157
01:03:24,001 --> 01:03:26,000
When you go to three, you get this curve.

1158
01:03:26,000 --> 01:03:28,000
When you go to four, you get this curve.

1159
01:03:28,001 --> 01:03:31,001
And the top curve there is 20,
but it kind of turns into a

1160
01:03:31,001 --> 01:03:33,000
triangle as you go off to many neighbors.

1161
01:03:33,000 --> 01:03:38,001
And I'm thinking that, you know, it's
probably around six neighbors or something.

1162
01:03:39,000 --> 01:03:40,000
So somewhere in here.

1163
01:03:40,001 --> 01:03:44,001
Now the observation to make is, well,
all we really wanted to do was pick a

1164
01:03:44,001 --> 01:03:46,001
received duty cycle here,
because we're going to have

1165
01:03:46,001 --> 01:03:48,000
one master schedule for the whole system.

1166
01:03:49,000 --> 01:03:54,001
And notice that if we pick a value of 0.
3, or maybe 0.33 or something, somewhere

1167
01:03:54,001 --> 01:03:58,000
around right in here,
I think what I used in the rest of the

1168
01:03:58,000 --> 01:03:59,001
simulations is actually 0.3.

1169
01:03:59,001 --> 01:04:03,001
If you pick 0.
3, you can ask, well, how far down are we from

1170
01:04:03,001 --> 01:04:05,000
the optimum in this case?

1171
01:04:05,001 --> 01:04:10,000
And we're already up to 20% here,
so we're only at about 0.3.

1172
01:04:10,001 --> 01:04:15,000
So we're only, we're more or less about
as good as we could have gotten, even

1173
01:04:15,000 --> 01:04:17,001
though we're only at 0.3 instead of 0.5.

1174
01:04:18,000 --> 01:04:20,000
And in the two case, well, we're right here.

1175
01:04:20,000 --> 01:04:21,001
Well, that's pretty close to the top two.

1176
01:04:21,001 --> 01:04:23,000
And in the three case, you're pretty close.

1177
01:04:23,000 --> 01:04:23,001
That's about right here.

1178
01:04:24,000 --> 01:04:25,000
And the four case, you're pretty right here.

1179
01:04:25,001 --> 01:04:27,001
And if you have to get around six,
you're about the right place.

1180
01:04:28,000 --> 01:04:31,000
So it doesn't really matter
much what you pick down here.

1181
01:04:31,000 --> 01:04:35,001
Something around 0.3 gets you pretty
close to optimal in all the cases.

1182
01:04:37,000 --> 01:04:38,001
And that's the received duty cycle, and

1183
01:04:38,001 --> 01:04:40,000
the transmit duty cycle is whatever's left.

1184
01:04:40,001 --> 01:04:43,000
So you listen about 30% of the time, and you

1185
01:04:43,000 --> 01:04:45,000
schedule your hand to your neighbors.

1186
01:04:46,000 --> 01:04:48,001
The schedule of your hand to your neighbors
is really a promise of when you're

1187
01:04:48,001 --> 01:04:50,001
going to be listening,
because that's all they really care about.

1188
01:04:50,001 --> 01:04:56,000
So you have about a third of, or 0.
30% of the time slots

1189
01:04:56,000 --> 01:04:57,001
in the future scheduled as received slots.

1190
01:04:58,001 --> 01:05:02,001
Now, of course,
there's also a notion here of clusters

1191
01:05:02,001 --> 01:05:04,000
of overlap neighbors, right?

1192
01:05:05,000 --> 01:05:07,001
You might have six neighbors here,
and the next six neighbors

1193
01:05:07,001 --> 01:05:09,001
here partially overlap, right?

1194
01:05:10,001 --> 01:05:12,000
Each station has their own view of the net.

1195
01:05:12,001 --> 01:05:13,000
Right.

1196
01:05:16,001 --> 01:05:17,000
So...

1197
01:05:23,001 --> 01:05:24,001
...perhaps, yes.

1198
01:05:25,001 --> 01:05:28,001
You may not actually turn your transmitter
on unless you actually have something

1199
01:05:28,001 --> 01:05:31,000
to send, and that depends on the traffic load,
and that's sort of a...

1200
01:05:35,000 --> 01:05:35,000
...what?

1201
01:05:37,000 --> 01:05:38,000
Vertical axis is...

1202
01:05:38,000 --> 01:05:41,001
Is the fraction of time that you
can spend transmitting. 100%,

1203
01:05:41,001 --> 01:05:43,000
80%, 60%, 40%, 20%.

1204
01:05:44,000 --> 01:05:47,001
The fraction of time you can spend
transmitting and do something useful.

1205
01:05:48,000 --> 01:05:50,001
I mean, the fraction of time you can spend
transmitting is just a line that runs

1206
01:05:50,001 --> 01:05:52,001
from this one up to this one.

1207
01:05:52,001 --> 01:05:55,000
One, you know, one minus P.

1208
01:05:56,000 --> 01:05:57,001
And that's why there's this limit here.

1209
01:05:57,001 --> 01:06:01,000
But it's not useful unless you've got at
least one neighbor listening, and that's

1210
01:06:01,000 --> 01:06:03,001
the difference between this line
and any particular curve we're on.

1211
01:06:09,000 --> 01:06:10,001
So, it's actually kind of weird.

1212
01:06:10,001 --> 01:06:13,000
I'm suggesting that you can spend more
than half of the time transmitting

1213
01:06:13,001 --> 01:06:15,000
and less than half the time listening.

1214
01:06:17,000 --> 01:06:18,001
More than half of the time...

1215
01:06:19,000 --> 01:06:20,001
The closer you can get.

1216
01:06:20,001 --> 01:06:22,000
Yeah, you can actually...

1217
01:06:22,000 --> 01:06:26,001
If you have six neighbors, two, three,
four, five, six, and at around 3% duty

1218
01:06:26,001 --> 01:06:30,000
cycle, you can actually have
throughputs that are 60% duty cycles.

1219
01:06:31,000 --> 01:06:31,001
And that's kind of weird.

1220
01:06:31,001 --> 01:06:35,000
You're spending 60% of the time transmitting
and only 40% of the time listening.

1221
01:06:35,000 --> 01:06:36,000
How can that possibly be?

1222
01:06:37,000 --> 01:06:38,001
Remember I said multiple of
these spreading channels.

1223
01:06:39,001 --> 01:06:41,001
When you're sending,
you're only sending one packet.

1224
01:06:44,000 --> 01:06:47,000
But when you're listening,
you might be capturing multiple at a time.

1225
01:06:47,001 --> 01:06:50,000
That's similar to one of the
things that a few years ago

1226
01:06:50,000 --> 01:06:52,000
was someone was proposing having that.

1227
01:06:52,000 --> 01:06:53,000
In case of your router.

1228
01:06:54,000 --> 01:06:57,001
And having one transmitter
and multiple receivers.

1229
01:06:58,001 --> 01:06:59,001
That was it.

1230
01:07:00,001 --> 01:07:00,001
That was it.

1231
01:07:00,001 --> 01:07:01,001
I don't know what they called it.

1232
01:07:02,000 --> 01:07:02,001
I'll put the transport on.

1233
01:07:03,000 --> 01:07:03,001
[...]

1234
01:07:03,001 --> 01:07:04,000
We'll have to roll area.

1235
01:07:05,000 --> 01:07:05,001
Yeah.

1236
01:07:06,000 --> 01:07:06,001
Yeah, that's it.

1237
01:07:06,001 --> 01:07:07,001
What year was that?

1238
01:07:07,001 --> 01:07:08,000
I haven't seen that.

1239
01:07:08,001 --> 01:07:08,001
[...]

1240
01:07:11,001 --> 01:07:16,000
I just, at Dayton,
I bought the whole proceedings for everything

1241
01:07:16,000 --> 01:07:17,001
from back to the first ECC.

1242
01:07:18,000 --> 01:07:20,000
And I didn't get up to that far.

1243
01:07:20,000 --> 01:07:22,000
I've only looked at the first five years.

1244
01:07:23,000 --> 01:07:24,001
We actually built a radio.

1245
01:07:26,000 --> 01:07:31,000
In fact, the Naval Research Lab actually
built a network similar to that, using

1246
01:07:31,000 --> 01:07:33,000
21 a bit per second.

1247
01:07:33,000 --> 01:07:37,000
[...] I think the channel
went UHF tax that and on HF.

1248
01:07:38,001 --> 01:07:43,000
And they actually were doing
that low rate of voice at like

1249
01:07:43,000 --> 01:07:45,000
400 or 600 bits per second.

1250
01:07:45,000 --> 01:07:46,000
Was it a single channel?

1251
01:07:46,001 --> 01:07:47,000
[...]

1252
01:07:47,000 --> 01:07:47,001
It was a...

1253
01:07:47,001 --> 01:07:48,001
Single channel system?

1254
01:07:50,000 --> 01:07:50,000
Multiple, yeah.

1255
01:07:50,001 --> 01:07:53,001
Multiple, it was,
each station had one transmitter.

1256
01:07:54,000 --> 01:07:55,000
It was on the air continuously.

1257
01:07:56,000 --> 01:07:59,001
It had like a router and you had
multiple receivers feeding it in.

1258
01:07:59,001 --> 01:08:01,000
Must have been a multiple channel system.

1259
01:08:01,000 --> 01:08:02,000
[...] throughput system.

1260
01:08:03,000 --> 01:08:06,001
Yeah, that's a different,
that's a whole different design place.

1261
01:08:06,001 --> 01:08:08,001
But it's not still because of what we're using

1262
01:08:08,001 --> 01:08:10,000
in current division instead of frequency.

1263
01:08:10,001 --> 01:08:11,000
[...]

1264
01:08:12,001 --> 01:08:14,000
I'll have to think about that.

1265
01:08:15,000 --> 01:08:17,000
It's actually, I think it's equivalent.

1266
01:08:21,000 --> 01:08:21,001
So...

1267
01:08:23,001 --> 01:08:24,000
So...

1268
01:08:24,001 --> 01:08:26,000
Let's return to this side briefly.

1269
01:08:26,001 --> 01:08:30,000
I sort of told you,
I hope a little bit more of this is...

1270
01:08:31,000 --> 01:08:34,000
Avoid remaining collisions by scheduling.

1271
01:08:34,000 --> 01:08:36,000
I just told you about the collision,
the scheduling method.

1272
01:08:37,001 --> 01:08:39,001
And I'm going to hand wave
the rest of these issues

1273
01:08:39,001 --> 01:08:41,000
and start showing you simulation results.

1274
01:08:42,000 --> 01:08:48,000
I sort of wrote an ad hoc special purpose
simulator to simulate the noise levels

1275
01:08:48,000 --> 01:08:50,000
or power levels in this kind of system.

1276
01:08:52,000 --> 01:08:52,001
And...

1277
01:08:53,000 --> 01:08:58,001
I wanted to see if, you know, doing the
scheduling method and with a whole bunch

1278
01:08:58,001 --> 01:09:02,001
of stations and actually compute routes
and actually send packets around whether

1279
01:09:02,001 --> 01:09:06,000
my signal to noise ratio
analysis actually works when you

1280
01:09:06,000 --> 01:09:07,001
actually have real stations out there.

1281
01:09:07,001 --> 01:09:09,000
So it's sort of a power level simulator.

1282
01:09:09,001 --> 01:09:14,000
It simulates the transfer of packets and
transmitter power levels and actually

1283
01:09:14,000 --> 01:09:18,000
records at the receiver, you know,
the power contribution by that transmitter.

1284
01:09:18,001 --> 01:09:20,001
And then when the transmitter shuts off,
it goes to the receiver and subtracts

1285
01:09:20,001 --> 01:09:23,000
out its contribution to the
noise without receiver.

1286
01:09:23,000 --> 01:09:29,001
So at every receiver as a function of time,
I've got what is the received total

1287
01:09:29,001 --> 01:09:31,001
power level in the channel.

1288
01:09:31,001 --> 01:09:35,000
Sort of before, you know, somebody asked
Matt yesterday in his talk, do you mean

1289
01:09:35,000 --> 01:09:37,000
like a received signal strength indicator?

1290
01:09:37,001 --> 01:09:40,001
Well, it'd be like sort of the before
de-spreading received signal strength

1291
01:09:40,001 --> 01:09:42,000
indicator is what I computed.

1292
01:09:42,000 --> 01:09:45,001
And then if you want to signal the
noise ratio for a given transfer, this

1293
01:09:45,001 --> 01:09:50,000
barrier, you know, this summation,
this sort of add in and then subtract out that

1294
01:09:50,000 --> 01:09:52,000
it is the station is sort of the total power.

1295
01:09:52,000 --> 01:09:55,000
And if you want to signal to noise ratio,
you have to again subtract out your own

1296
01:09:55,000 --> 01:09:58,000
signal level because what,
if you just take the signal level to that, you

1297
01:09:58,000 --> 01:09:59,001
get the signal to signal plus noise ratio.

1298
01:10:02,000 --> 01:10:05,001
So I place stations randomly in
a square by just picking two

1299
01:10:05,001 --> 01:10:07,000
random numbers from the
random number generator.

1300
01:10:07,001 --> 01:10:10,000
I computed minimum energy routes.

1301
01:10:11,001 --> 01:10:14,001
And what I've drawn on this
picture are all the single hops

1302
01:10:14,001 --> 01:10:16,000
that any of the minimum energy routes use.

1303
01:10:17,000 --> 01:10:21,000
Now minimum energy routes have the property
that if you have a route from B to C,

1304
01:10:23,000 --> 01:10:28,000
and the route from A to C goes through B,
then the end of that route, once you

1305
01:10:28,000 --> 01:10:30,001
get to B, is the same as the route from B to C.

1306
01:10:31,001 --> 01:10:33,000
It's got to be that way.

1307
01:10:34,000 --> 01:10:39,000
So you can sort of think about, you know,
so all you need to remember, minimum

1308
01:10:39,000 --> 01:10:42,000
energy routes, all you need to remember
at any station is how to get there.

1309
01:10:42,000 --> 01:10:44,001
And then when the packet shows up,
that's the way you send it.

1310
01:10:44,001 --> 01:10:47,000
You don't have to send it a different
route depending on where it came

1311
01:10:47,000 --> 01:10:49,000
from, minimum energy route.

1312
01:10:49,000 --> 01:10:52,001
So I compute the routes,
put a routing table at each station,

1313
01:10:53,000 --> 01:10:55,000
and then started playing with the simulator.

1314
01:10:57,000 --> 01:11:01,001
As I said, the length, the edges on
this diagram between the random of the

1315
01:11:01,001 --> 01:11:03,000
playstations show where the

1316
01:11:03,000 --> 01:11:08,001
single hops are that are used by any route.

1317
01:11:10,000 --> 01:11:14,000
So I stared at some of these pictures,
didn't really learn anything.

1318
01:11:15,001 --> 01:11:19,000
Are you thinking of it in context of
like a metropolitan area network?

1319
01:11:19,001 --> 01:11:21,000
Absolutely,
it was like about a million stations.

1320
01:11:22,000 --> 01:11:23,000
There's a hundred in this picture.

1321
01:11:25,000 --> 01:11:26,000
I'll get to a thousand in a minute.

1322
01:11:27,001 --> 01:11:33,000
I would imagine these are traditional
methods for bridging that metropolitan

1323
01:11:33,000 --> 01:11:34,001
area network into a light area network.

1324
01:11:35,001 --> 01:11:37,000
Well, you'll have to have some probably...

1325
01:11:37,000 --> 01:11:39,000
That's going to be a collapse back on there.

1326
01:11:39,001 --> 01:11:43,000
Yeah, absolutely. Fiber, you know,
some long-distance provider that you'll pay

1327
01:11:43,000 --> 01:11:44,001
for service from to get somewhere.

1328
01:11:44,001 --> 01:11:45,001
Could you go back one price?

1329
01:11:47,000 --> 01:11:49,001
That was the diagram of the...

1330
01:11:50,001 --> 01:11:54,001
Like, for example, what I'm looking at
is the node right here, the center, the

1331
01:11:54,001 --> 01:11:57,000
55, the [...] that guy.

1332
01:11:57,001 --> 01:12:00,001
Now, if everybody wants to talk to everybody
equally, don't you have an infinite

1333
01:12:00,001 --> 01:12:02,001
amount of traffic going through every node?

1334
01:12:04,000 --> 01:12:05,000
Infinite going through?

1335
01:12:06,000 --> 01:12:07,000
I'll get to that.

1336
01:12:08,000 --> 01:12:10,000
Actually,
I'll get to that in a thousand stations.

1337
01:12:11,000 --> 01:12:12,001
There's actually more discussion about that

1338
01:12:12,001 --> 01:12:14,000
in the thesis I'm probably
not going to get into.

1339
01:12:18,001 --> 01:12:22,000
You're worrying about exactly
the right thing and...

1340
01:12:22,000 --> 01:12:23,000
I'm very excited.

1341
01:12:23,000 --> 01:12:26,001
And when I started looking at the simulation
results, exactly what you're worried

1342
01:12:26,001 --> 01:12:29,000
about was in fact a problem, and

1343
01:12:29,000 --> 01:12:30,001
I haven't completely figured
out how to solve it yet.

1344
01:12:31,000 --> 01:12:32,000
So you're on the right track.

1345
01:12:32,000 --> 01:12:35,000
The ball I'm posting is my neighbor's.
You got a fantastic system.

1346
01:12:35,001 --> 01:12:37,001
Well,
I'll wave my hands at the end of the talk.

1347
01:12:38,000 --> 01:12:40,000
I was going to...
maybe it may relate to the same question.

1348
01:12:40,000 --> 01:12:44,000
I was going to want to...
did you look at, say, varying the mix of

1349
01:12:44,000 --> 01:12:46,000
unicast versus multicast traffic?

1350
01:12:46,000 --> 01:12:48,000
I have...I have...

1351
01:12:48,000 --> 01:12:50,000
I forget what...

1352
01:12:53,000 --> 01:12:55,000
I'm going to read you the
last sentence in the last

1353
01:12:55,000 --> 01:12:56,001
chapter of my thesis, or the last two.

1354
01:12:59,001 --> 01:13:00,001
Just a...

1355
01:13:00,001 --> 01:13:03,001
That's the question I was wondering
whether multicast and...the

1356
01:13:03,001 --> 01:13:05,001
minimum range of routing still...

1357
01:13:05,001 --> 01:13:08,000
The last sentence was... ...
still the optimal range of routing.

1358
01:13:08,000 --> 01:13:11,001
You know, where I ended my thesis was I said,
after talking about a whole bunch

1359
01:13:11,001 --> 01:13:16,000
of issues, I said, these questions,
these questions again depend on routing and

1360
01:13:16,000 --> 01:13:19,000
demand for service, questions that are
difficult to determine before real users

1361
01:13:19,000 --> 01:13:21,000
are involved in the generation of traffic.

1362
01:13:22,000 --> 01:13:22,001
I don't know.

1363
01:13:25,001 --> 01:13:28,001
But I'll wave my hands at another
picture that might actually

1364
01:13:28,001 --> 01:13:30,000
be able to get you excited.

1365
01:13:30,000 --> 01:13:32,000
That's why I wanted to show
one of the notion on the

1366
01:13:32,000 --> 01:13:33,001
table here, and that's a deployment issue.

1367
01:13:36,001 --> 01:13:40,000
Obviously, there's got to be a
consideration for the distance between

1368
01:13:40,000 --> 01:13:42,000
your initial stations.

1369
01:13:42,001 --> 01:13:44,000
What's the...you copy a man...

1370
01:13:44,001 --> 01:13:45,000
Yeah, what is...

1371
01:13:45,000 --> 01:13:47,000
There's two different...
...fringence of that man.

1372
01:13:47,000 --> 01:13:50,000
If you only have two stations,
how far apart can they can be when you

1373
01:13:50,000 --> 01:13:52,000
initially put them up, and then you'll fill in.

1374
01:13:53,000 --> 01:13:53,001
Absolutely.

1375
01:13:55,001 --> 01:14:00,000
So, I think I said what was on this
slide before I put up the last slide.

1376
01:14:01,000 --> 01:14:03,001
But I explicitly reported the
contribution to the noise level at each

1377
01:14:03,001 --> 01:14:05,000
receiver for each transmission.

1378
01:14:08,000 --> 01:14:11,001
And I...by record, I mean I added
into the total there, and when the

1379
01:14:11,001 --> 01:14:13,000
transmitter ends, I subtract it out again.

1380
01:14:13,001 --> 01:14:15,001
So there's a...I've got a variable for each

1381
01:14:15,001 --> 01:14:17,000
station, which is the current received.

1382
01:14:18,000 --> 01:14:22,000
There's a vent queue driven for the traffic,
so you can simulate, you know...

1383
01:14:23,000 --> 01:14:25,000
I'm starting to send a packet,
and I know the packet

1384
01:14:25,000 --> 01:14:27,000
is this long, so it's going to...

1385
01:14:27,000 --> 01:14:29,001
The transmission is going to
end at this time in the future.

1386
01:14:30,000 --> 01:14:33,001
It schedules an event in the event queue,
which is a heat priority queue, so it

1387
01:14:33,001 --> 01:14:36,001
only takes long time to
extract an event out of it.

1388
01:14:37,000 --> 01:14:40,001
It's only logged in the size of the queue,
so it's a standard sort of

1389
01:14:40,001 --> 01:14:42,000
event driven network simulator.

1390
01:14:43,001 --> 01:14:49,001
And for each...when a station is
actually receiving a packet, it's also

1391
01:14:49,001 --> 01:14:55,001
got something which is taking sort of the
max of the noise level, so that you can

1392
01:14:55,001 --> 01:14:59,000
find the NIV of the signal to noise
ratio at the end of the reception of the

1393
01:14:59,000 --> 01:15:03,000
packet, and record what that was,
and decide whether or not

1394
01:15:03,000 --> 01:15:04,001
that packet reception was successful.

1395
01:15:06,000 --> 01:15:10,001
Now, in the simulations I'm
going to talk about today, I...

1396
01:15:12,001 --> 01:15:14,000
Do you need to piece that one more time?

1397
01:15:17,001 --> 01:15:23,001
The simulator has, at every station,
a variable which is the current noise total

1398
01:15:23,001 --> 01:15:25,001
power level coming from all the other stations.

1399
01:15:26,001 --> 01:15:29,001
Actually, I initialized that variable
with the thermal noise level, but I could

1400
01:15:29,001 --> 01:15:31,001
have initialized it with zero,
and I would have gotten the same result

1401
01:15:31,001 --> 01:15:34,000
in the simulations,
because that's far enough away.

1402
01:15:34,000 --> 01:15:41,000
So, that variable, I'm also...
during the reset, while a station is

1403
01:15:41,000 --> 01:15:46,000
receiving a packet, I record for that
packet what the max of this variable

1404
01:15:46,000 --> 01:15:48,000
was, which is the total power level.

1405
01:15:48,001 --> 01:15:52,000
So that when the event happens,
it marks the end of the reception of a

1406
01:15:52,000 --> 01:15:53,001
packet in the simulator.

1407
01:15:54,000 --> 01:15:58,000
I can then go look at what the max noise
level was during the reception of this

1408
01:15:58,000 --> 01:15:59,001
packet,
compute the signal level for that packet, and

1409
01:15:59,001 --> 01:16:01,001
that's the signal to noise
ratio for that packet.

1410
01:16:03,000 --> 01:16:06,000
That's the worst signal to
noise ratio experience during

1411
01:16:06,000 --> 01:16:08,001
the reception time of this packet.

1412
01:16:09,000 --> 01:16:13,000
Are you going to take that one step
further and actually make that variable

1413
01:16:13,000 --> 01:16:15,001
available in the real production unit, or...

1414
01:16:16,001 --> 01:16:18,000
Does that sound like a useful thing to do?

1415
01:16:18,001 --> 01:16:20,001
I don't know. I haven't fully thought

1416
01:16:20,001 --> 01:16:22,000
about exactly what a real production unit was.

1417
01:16:22,000 --> 01:16:25,000
This is for the purpose of understanding
whether my signal to noise ratio

1418
01:16:25,000 --> 01:16:27,001
calculations were correct earlier.

1419
01:16:28,000 --> 01:16:32,000
I have no particular designs,
plans to build any particular piece of hardware

1420
01:16:32,000 --> 01:16:33,001
at this moment,
so I would love to build some hardware.

1421
01:16:34,001 --> 01:16:40,000
Yeah. So, now I call it...
there's all sorts of constants.

1422
01:16:40,001 --> 01:16:42,001
I actually went through the simulator and tried

1423
01:16:42,001 --> 01:16:44,000
to actually get all the constants right.

1424
01:16:44,000 --> 01:16:49,000
I assumed isotropic antennas and actually
figured out what the propagation

1425
01:16:49,000 --> 01:16:50,001
is based on isotropic antennas.

1426
01:16:51,000 --> 01:16:54,000
And so if you put it, you know, have
antennas of some, you know, on the antennas

1427
01:16:54,000 --> 01:16:59,000
with like three or five dB of gain,
that would be different. I plugged in all the

1428
01:16:59,000 --> 01:17:01,001
numbers for the 902 to 928 MHz band.

1429
01:17:01,001 --> 01:17:04,000
It assumed it's 26 MHz wide
and we're using all of that.

1430
01:17:04,001 --> 01:17:09,000
So the sort of noise bandwidth
before the spreading is 26 MHz wide,

1431
01:17:09,001 --> 01:17:15,001
which is the signal to noise
ratio computation that I make.

1432
01:17:17,000 --> 01:17:21,001
And I assumed...now the scheduling method,
I assumed 64 slots per second.

1433
01:17:21,001 --> 01:17:26,000
And actually I skipped over how you actually
find the place a packet can sit in.

1434
01:17:26,000 --> 01:17:31,000
And I assumed fault small or
fixed by packets of some size

1435
01:17:31,000 --> 01:17:33,000
without actually specifying a data rate.

1436
01:17:33,000 --> 01:17:36,000
So I'm actually just saying a time and
the data rate we can get depends on the

1437
01:17:36,000 --> 01:17:39,000
processing gain that we need, which...
and that'll determine

1438
01:17:39,000 --> 01:17:41,000
how many bits you can send in a packet.

1439
01:17:42,001 --> 01:17:44,000
But there are these things called sub slots.

1440
01:17:44,000 --> 01:17:50,000
So this is 1024 sub slots per slot,
which means the packet is at most about

1441
01:17:50,000 --> 01:17:53,000
a millisecond long.

1442
01:17:53,001 --> 01:17:57,001
And a target power control of minus 120
dBw when I have power control turned on.

1443
01:18:00,001 --> 01:18:04,001
And I play with a bunch of stuff.
The only simulations I wrote up in the thesis,

1444
01:18:05,001 --> 01:18:09,001
because they were the only ones that
really informed anything, was this worst,

1445
01:18:09,001 --> 01:18:11,001
near worst case traffic
load that I came up with.

1446
01:18:11,001 --> 01:18:15,001
I computed the routes to handle
general traffic. I really have no idea

1447
01:18:15,001 --> 01:18:17,000
what traffic's going to be put on this.

1448
01:18:17,001 --> 01:18:19,001
But what I'm concerned about is,
well, if it got really

1449
01:18:19,001 --> 01:18:21,001
bad, how bad would all the interference be?

1450
01:18:22,000 --> 01:18:24,001
And what I wanted to get is I wanted
to get every station sort of sending

1451
01:18:24,001 --> 01:18:26,000
to all its neighbors as much as possible.

1452
01:18:27,000 --> 01:18:32,000
So coming back to this diagram of the
network and the edges connecting them, if

1453
01:18:32,000 --> 01:18:36,000
you imagine any station here,
and each station here has like four or five

1454
01:18:36,000 --> 01:18:38,000
edges off to some nearby neighbor,

1455
01:18:41,001 --> 01:18:48,001
then for each station in the net,
pick one station, I schedule the packet to

1456
01:18:48,001 --> 01:18:52,001
each of its neighbors when
the simulator initializes.

1457
01:18:53,000 --> 01:18:55,000
So every station starts out with one
packet for each of its neighbors.

1458
01:18:55,001 --> 01:18:59,001
At the start time for transmission
of each packet, I scheduled

1459
01:18:59,001 --> 01:19:01,001
another packet for that same neighbor.

1460
01:19:01,001 --> 01:19:03,001
And then it goes and finds a place in the

1461
01:19:03,001 --> 01:19:05,000
pseudo-random schedules where it can go.

1462
01:19:05,000 --> 01:19:10,000
So every station is always transmitting
at any time it has a neighbor listening.

1463
01:19:11,001 --> 01:19:16,001
It's completely loaded. No opportunity.
Now, the power level, if you turn on

1464
01:19:16,001 --> 01:19:19,000
power control, you're not always
transmitting to your furthest neighbor.

1465
01:19:19,001 --> 01:19:24,000
So it's not entirely worst case.
If you've got a, you know, it's sort of, you're

1466
01:19:24,000 --> 01:19:26,001
always transmitting to some neighbor,
but if some of your neighbors are closer

1467
01:19:26,001 --> 01:19:28,001
than others,
then your transmitter is not at the highest

1468
01:19:28,001 --> 01:19:30,000
level you ever run it at all the time.

1469
01:19:31,001 --> 01:19:33,001
Maybe I should do that simulation as well.

1470
01:19:33,001 --> 01:19:39,000
Does that mean like, you know,
somebody's transmitting to your [...] for it?

1471
01:19:40,000 --> 01:19:42,000
No, the schedules are completely fixed.
You have to fit the

1472
01:19:42,000 --> 01:19:43,001
packets in the schedules that are there.

1473
01:19:44,000 --> 01:19:51,000
So you have to, oh, okay.
You can't, you can't send, you got 120 8-byte

1474
01:19:51,000 --> 01:19:55,001
packets, but then, and there's a few spots,
80 bytes long, you got an 80 byte

1475
01:19:55,001 --> 01:19:57,000
period, and wait for the next time.

1476
01:19:57,000 --> 01:20:01,000
Well, yeah, the way I probably, I don't know.
Maybe the way I build the system

1477
01:20:01,000 --> 01:20:02,001
is I'd have a small, fixed-size cell.

1478
01:20:03,001 --> 01:20:08,001
Which is maybe one-fourth of a slot time.
I think that's probably about right.

1479
01:20:10,000 --> 01:20:14,000
And towards the end of the panel for the
chapter in the thesis are some formulas.

1480
01:20:14,000 --> 01:20:18,001
You can start plugging in what sizes of
different things you want and see what

1481
01:20:18,001 --> 01:20:20,001
efficiencies you get in
various pieces of the system.

1482
01:20:21,001 --> 01:20:27,000
Now here, what I have on this plot is, let me

1483
01:20:27,000 --> 01:20:29,000
think, which one, yeah, this one, okay.

1484
01:20:29,000 --> 01:20:36,000
This is, I just picked one of the stations
in the net, actually, station 29,

1485
01:20:36,001 --> 01:20:40,001
which is this station that has
three neighbors down here.

1486
01:20:40,001 --> 01:20:43,000
One of which is kind of nearby
and two which are further away.

1487
01:20:45,001 --> 01:20:49,001
And I, what I have is here is a
plot of received power level.

1488
01:20:50,001 --> 01:20:55,000
The thermal noise for 26 megahertz
wide is around minus 150 dBw.

1489
01:20:57,001 --> 01:21:03,000
That's Boltzmann's constant times the
temperature, 290 times 26 megahertz.

1490
01:21:04,000 --> 01:21:08,001
And so, you know, the, the,
we're about 50 dB away from that, just

1491
01:21:08,001 --> 01:21:10,000
from the self-interference in the system.

1492
01:21:11,000 --> 01:21:14,000
What I've got on this axis is time,
horizontal axis is time, and the vertical

1493
01:21:14,000 --> 01:21:16,001
axis here is received power level.

1494
01:21:17,000 --> 01:21:22,001
This is minus 60 dBw or a micro-lot,
nano-lot, pico-lot.

1495
01:21:26,000 --> 01:21:28,001
And there's another little
bit of hand waving here.

1496
01:21:31,000 --> 01:21:34,001
These, these tall spikes, when you see
received power levels up around the micro

1497
01:21:34,001 --> 01:21:37,001
-lot, that's the station's own transmitter.

1498
01:21:39,000 --> 01:21:42,001
I forget the actual power level I had the
transmitter running at, but it was on

1499
01:21:42,001 --> 01:21:46,000
the order of somewhere between
10 and 100 milliwatts, I think.

1500
01:21:47,000 --> 01:21:51,000
Actually, it's exactly 40 dB up from
this because even though I think you can

1501
01:21:51,000 --> 01:21:55,000
only build isolators or duplexors
with about 20 dB of isolation.

1502
01:21:56,001 --> 01:21:58,000
Circulators, sometimes people call them.

1503
01:21:58,000 --> 01:21:59,001
They're duplexors to telephony people.

1504
01:22:01,000 --> 01:22:01,001
That's right.

1505
01:22:05,000 --> 01:22:09,000
I first plugged in 20 dBw and
all my simulations diverged.

1506
01:22:10,000 --> 01:22:12,001
And the problem is,
is I initialize this variable.

1507
01:22:12,001 --> 01:22:19,000
I mean, this, but what happened is,
is you, because a

1508
01:22:19,000 --> 01:22:25,001
round off error in IEEE floating point
standard doubles, which only have

1509
01:22:25,001 --> 01:22:30,000
some,
there's a footnote in the thesis about this.

1510
01:22:30,001 --> 01:22:33,000
There's only so many bits of mantissa.

1511
01:22:33,001 --> 01:22:37,001
And if you take a big thing,
if you have a small thing and you add in a big

1512
01:22:37,001 --> 01:22:41,000
thing, and this just tracks the
big thing out again, you might

1513
01:22:41,000 --> 01:22:43,000
not get exactly what you started with.

1514
01:22:43,001 --> 01:22:46,000
And if that starts creeping, then it runs away.

1515
01:22:47,000 --> 01:22:52,000
And I had to, there's only about 150
dB of dynamic range in the mantissa

1516
01:22:52,000 --> 01:22:54,000
of an IEEE floating point double.

1517
01:22:55,000 --> 01:22:56,000
Or 64 bits, yeah.

1518
01:22:56,001 --> 01:23:01,000
So if you want to add signals that are
power levels that are more than 150 dB, or

1519
01:23:01,000 --> 01:23:05,001
that are close that approach 150 dB difference,
then you need,

1520
01:23:06,000 --> 01:23:09,000
you need something bigger than IEEE double.

1521
01:23:09,000 --> 01:23:12,001
And the way I solve this problem,
so I can run all my simulations, is I wave my

1522
01:23:12,001 --> 01:23:16,001
hands and let's say, okay, let's say the
circulator gives us 40 dB of isolation.

1523
01:23:17,000 --> 01:23:22,001
So these tall spikes here are us hearing
our own transmitter through a 40 dB

1524
01:23:22,001 --> 01:23:25,001
circulator,
a circulator giving us 40 dB of isolation.

1525
01:23:26,000 --> 01:23:27,001
Really, these would be taller.

1526
01:23:29,001 --> 01:23:30,001
What? How do you do that?

1527
01:23:34,001 --> 01:23:35,000
Yeah.

1528
01:23:36,000 --> 01:23:36,001
Yeah, yeah.

1529
01:23:37,001 --> 01:23:39,001
So, so that's what these tall spikes are.

1530
01:23:40,000 --> 01:23:43,000
So the transmitter actually radiated,
you know, the power we're sending up the

1531
01:23:43,000 --> 01:23:46,001
antenna is 40 dB above this,
because we're listening to it through,

1532
01:23:47,000 --> 01:23:48,001
effectively, a 40 dB of attenuation.

1533
01:23:49,000 --> 01:23:55,000
And, and notice that's about 40 dB above
the received power levels, the total

1534
01:23:55,000 --> 01:23:57,001
received power level as well,
and about 50 dB above the actual

1535
01:23:57,001 --> 01:23:59,001
receptions of the signal levels of the packets.

1536
01:23:59,001 --> 01:24:06,001
What I've drawn in here is the, the,
the signal level of individual packets.

1537
01:24:06,001 --> 01:24:10,001
So this is the noise, and the noise goes,
I mean, the noise includes all the

1538
01:24:10,001 --> 01:24:14,000
interference,
and it's tall when it's up, minus 60

1539
01:24:14,000 --> 01:24:16,000
dBw when our own transmitter's on.

1540
01:24:16,000 --> 01:24:18,001
That's around minus 100 dBw
when our own transmitter's off.

1541
01:24:20,000 --> 01:24:24,000
And the receptions are here.
So the distance from these receptions up to the

1542
01:24:24,000 --> 01:24:29,001
noise level is kind of the signal
to noise plus signal ratio.

1543
01:24:33,001 --> 01:24:37,001
And you can't really learn much from this plot,
except I can wave at my hands

1544
01:24:37,001 --> 01:24:39,000
at it while I talk about
how the simulator works.

1545
01:24:40,000 --> 01:24:41,000
And, show up for you here.

1546
01:24:41,001 --> 01:24:47,000
There's no, yeah, more or less,
except this is signal to signal

1547
01:24:47,000 --> 01:24:48,001
plus noise ratio, not signal to noise ratio.

1548
01:24:49,001 --> 01:24:52,000
Which, if, if this were further apart,
they're kind of the

1549
01:24:52,000 --> 01:24:54,000
same thing. They approach the same thing.

1550
01:24:54,000 --> 01:24:57,001
Now, notice the receptions come at a
different power level. That's because in

1551
01:24:57,001 --> 01:25:00,000
this simulation I was running a fixed

1552
01:25:00,000 --> 01:25:01,001
transmitter power level throughout the system.

1553
01:25:01,001 --> 01:25:03,000
All, no power control.

1554
01:25:05,001 --> 01:25:10,000
And I have,
this is when you're on transmitter on.

1555
01:25:10,000 --> 01:25:15,001
What this tall is, when the noise
interference level pops up about 10 dB,

1556
01:25:16,001 --> 01:25:21,001
that's due to that really nearby
neighbor who's got his transmitter on.

1557
01:25:22,000 --> 01:25:25,001
And he kind of pops up out of the noise.
And if you had a reception down

1558
01:25:25,001 --> 01:25:27,001
here, you suddenly, you have a problem.

1559
01:25:28,000 --> 01:25:33,001
And I'll come back to that later.
So that's, that should more or less explain,

1560
01:25:33,001 --> 01:25:35,000
you know. Exactly.

1561
01:25:35,001 --> 01:25:37,000
Some people call that a near-power problem.

1562
01:25:44,000 --> 01:25:44,001
[...]

1563
01:25:54,001 --> 01:26:00,000
Yeah, but I, yeah, that, I believe you.
Absolutely. I think that would do it.

1564
01:26:00,000 --> 01:26:03,000
And you'd be doing a random,
instead of always being biased

1565
01:26:03,000 --> 01:26:06,000
wrong, you'd be doing a random walk.

1566
01:26:06,001 --> 01:26:10,000
Which hopefully wouldn't walk too far.
But this is an IEEE floating point unit in

1567
01:26:10,000 --> 01:26:12,001
my microprocessor. I don't know if I,
I don't know if I can tweak that.

1568
01:26:13,000 --> 01:26:17,001
I'm just writing code in C and running it on a,
I was actually running on a,

1569
01:26:17,001 --> 01:26:19,000
on a MIPSAR 3000 at the time.

1570
01:26:21,001 --> 01:26:24,001
And I, I think it had mode bits,
but I don't think you could get at them from

1571
01:26:24,001 --> 01:26:26,001
user mode. So I'd have to hack the kernel.

1572
01:26:33,000 --> 01:26:37,001
Yeah, well I found another way around this.

1573
01:26:39,001 --> 01:26:41,000
You should have talked to me four years ago.

1574
01:26:42,001 --> 01:26:45,000
I'm still not sure if I would have been
able to do it in the computer I had.

1575
01:26:45,001 --> 01:26:48,000
So, interesting idea.

1576
01:26:48,001 --> 01:26:52,000
So this was, like, this is the flight I
had up just a moment ago. This is the

1577
01:26:52,000 --> 01:26:53,001
without power control.

1578
01:26:53,001 --> 01:26:56,000
Here I turned on power control, and this

1579
01:26:56,000 --> 01:26:58,000
station was sending to three other stations.

1580
01:26:58,000 --> 01:27:00,000
And you can see that our
transmitter power level

1581
01:27:00,000 --> 01:27:02,000
goes between three different values.

1582
01:27:02,001 --> 01:27:06,000
This is when we're sending to that guy
who's really nearby. This is sending

1583
01:27:06,000 --> 01:27:07,001
to one of the other two visits further away.

1584
01:27:08,000 --> 01:27:11,000
And you can see this is quite different
than power control. And the power control

1585
01:27:11,000 --> 01:27:15,000
throughout the whole network in this
simulation, conspired to always deliver

1586
01:27:15,000 --> 01:27:19,001
minus 120 dBw to the receiver.
So all the receptions are exactly this level.

1587
01:27:21,001 --> 01:27:24,001
Just to illustrate what I mean by power
control, you can sort of see it here.

1588
01:27:25,000 --> 01:27:27,001
Now you can't really learn anything.
You know, these

1589
01:27:27,001 --> 01:27:29,001
two plots look a little different.

1590
01:27:30,000 --> 01:27:34,001
You know, with power control it was a
little lower, but I could have conspired to

1591
01:27:34,001 --> 01:27:37,000
fix the transmitter power level so
that they kind of look the same.

1592
01:27:38,000 --> 01:27:41,000
You know, there's really not much
to learn by comparing those two.

1593
01:27:42,001 --> 01:27:46,000
What I've done here, now this is with,
again with power control, this is from the

1594
01:27:46,000 --> 01:27:47,001
simulation without power control.

1595
01:27:48,001 --> 01:27:52,000
And what I've done on this, now remember this,

1596
01:27:53,000 --> 01:27:54,001
I told you what the traffic load was.

1597
01:27:54,001 --> 01:27:59,000
I was just loading up to send to nearby
neighbors. So all, every station is

1598
01:27:59,000 --> 01:28:02,000
constantly, they'll be sending to a nearby
neighbor throughout the entire NAT.

1599
01:28:03,000 --> 01:28:09,000
And so what I did for all my simulations,
I, what I have on this axis, on the

1600
01:28:09,000 --> 01:28:12,000
horizontal axis,
is signal to noise ratio in dustables.

1601
01:28:12,000 --> 01:28:18,001
There's 20 dB, 10 dB, 0 dB, minus 10 dB,
minus 20 dB. So this is the

1602
01:28:18,001 --> 01:28:23,000
one, one in 100,
one part in 100 signal to noise ratio.

1603
01:28:24,001 --> 01:28:28,001
I refuse to say negative signal to noise ratio.
It's always a positive signal to

1604
01:28:28,001 --> 01:28:30,001
noise ratio. It's just the log of
the signal to noise ratio that

1605
01:28:30,001 --> 01:28:32,000
is negative when you're in dB.

1606
01:28:33,001 --> 01:28:38,000
So this is what sort of distribution
you get of signal, worst case signal to

1607
01:28:38,000 --> 01:28:41,000
noise ratios during receptions of
packets without power control turned on.

1608
01:28:41,000 --> 01:28:46,000
And you see this kind of blob. And if you
picked, the way to think about this is

1609
01:28:46,000 --> 01:28:52,000
that if you picked a processing gain
of say 23 dB, and you needed 5 dB of

1610
01:28:52,000 --> 01:28:55,000
headroom,
then all of the packets to the right of here

1611
01:28:55,000 --> 01:28:57,001
in this distribution would have been received.

1612
01:28:58,000 --> 01:29:04,000
And only these few down here might not
have been received. So this kind of sort

1613
01:29:04,000 --> 01:29:07,001
of, even this without power
control sort of confirms,

1614
01:29:08,000 --> 01:29:09,001
you know,
more or less what I was thinking about.

1615
01:29:09,001 --> 01:29:14,000
You know, this thing ought to work
from my analysis earlier. And I'll

1616
01:29:14,000 --> 01:29:15,001
come back and actually do a comparison later.

1617
01:29:16,000 --> 01:29:21,000
What I want to do right now is actually
turn on power control and it tightens up

1618
01:29:21,000 --> 01:29:26,000
the distribution.
This plot is at the same scale as the previous.

1619
01:29:26,001 --> 01:29:29,001
What it's done is it's

1620
01:29:29,001 --> 01:29:36,000
taken what was in this distribution,
I can put them over

1621
01:29:36,000 --> 01:29:38,000
each other, and scrunched them together.

1622
01:29:38,000 --> 01:29:41,001
And it sort of makes sense because what
it's doing is it's fixing the signal

1623
01:29:41,001 --> 01:29:45,000
level everywhere. And we already think
that because there's so many stations, the

1624
01:29:45,000 --> 01:29:47,000
noise level is sort of what
it is in a large network.

1625
01:29:47,000 --> 01:29:52,000
There's only 100 stations in the simulation,
but the power control, that's what

1626
01:29:52,000 --> 01:29:57,001
power control does for you. It sort of
eliminates the very high signal to noise

1627
01:29:57,001 --> 01:30:01,000
ratios because when that would have happened,
you turn your power down.

1628
01:30:01,000 --> 01:30:07,001
And because people turn their power down,
it reduces the number of

1629
01:30:07,001 --> 01:30:09,000
very low signal to noise ratio

1630
01:30:09,000 --> 01:30:11,000
receptions and sort of brings
them closer to the middle.

1631
01:30:11,001 --> 01:30:16,001
Now what I want to do is actually I'm
going to zoom in on this picture. This is

1632
01:30:16,001 --> 01:30:21,000
with power control turned on.
And I'm going to look at this tail here.

1633
01:30:21,000 --> 01:30:24,001
So I'm going to zoom in on this tail.
I'm going to sort of blow the scale up of

1634
01:30:24,001 --> 01:30:31,000
this plot with my next plot. And I actually
moved zero up, the zero level up a

1635
01:30:31,000 --> 01:30:34,000
little bit off the y-axis so we
could see this more clearly.

1636
01:30:34,001 --> 01:30:40,000
And this tail goes out for a ways and
there's some very low, there's some packets

1637
01:30:40,000 --> 01:30:44,000
that came in with fairly low
signal to noise ratio. And what

1638
01:30:44,000 --> 01:30:47,001
happened? It's the near far problem.

1639
01:30:50,000 --> 01:30:55,000
And when I explained today my scheduling
method, I just talked about well we'll

1640
01:30:55,000 --> 01:31:02,000
schedule to avoid, you know,
my neighbors will know when they can send to me

1641
01:31:02,000 --> 01:31:04,001
because they'll know when I'm not transmitting.

1642
01:31:04,001 --> 01:31:09,000
So we don't have, you know, extreme,
so there were no extremely low signal to

1643
01:31:09,000 --> 01:31:13,000
noise ratios. That never happened because
the scheduling method was working. We

1644
01:31:13,000 --> 01:31:17,001
didn't see any minus 100 or minus
80 dB signal to noise ratios.

1645
01:31:17,001 --> 01:31:19,001
Or it would be closer to minus 40.

1646
01:31:19,001 --> 01:31:24,001
But, time's running along.

1647
01:31:28,000 --> 01:31:32,000
But there was this near far problem.
Now, the scheduling method is fairly easy to

1648
01:31:32,000 --> 01:31:36,001
extend since I know what your schedule
is and I know that I'm close to you.

1649
01:31:37,000 --> 01:31:38,001
And I know what my schedule is.

1650
01:31:39,000 --> 01:31:43,000
It's fairly easy for me, and I know what
the path gain is. I'm assuming we've

1651
01:31:43,000 --> 01:31:48,000
done enough for that for power control.
So if I'm going to send over there and

1652
01:31:48,000 --> 01:31:52,001
I'm nearby some other receiver and I know
what his schedule is, when I'm finding

1653
01:31:52,001 --> 01:31:56,000
the place that I'm going to send
this packet to him, I can look at

1654
01:31:56,000 --> 01:31:57,001
his received schedule as well.

1655
01:31:58,001 --> 01:32:04,001
And choose to send when I, if I'm going
to send a packet far away, I'm going,

1656
01:32:04,001 --> 01:32:08,000
what I can do is I can look at,
I need to find a time when that's a received

1657
01:32:08,000 --> 01:32:12,001
slot, I've got a transmit slot,
and he's in a transmit slot.

1658
01:32:14,000 --> 01:32:19,001
Exactly. Now, does this really constrain
things much? Well, it's only about

1659
01:32:19,001 --> 01:32:26,001
20% of the time that I can send to him.
What fraction of, and

1660
01:32:26,001 --> 01:32:31,000
so there's a given schedule that I can
send across the room. You know, I can send

1661
01:32:31,000 --> 01:32:36,000
if somebody is fairly distant. What
fraction of his, so, and all the schedules,

1662
01:32:36,000 --> 01:32:42,000
what fraction of his slots are transmit slots?
70%, a majority. So it actually

1663
01:32:42,000 --> 01:32:45,000
doesn't constrain you that much.
And it's only when you're sitting right on top

1664
01:32:45,000 --> 01:32:50,001
of somebody. In fact, it never comes into
play unless there's a factor of about

1665
01:32:50,001 --> 01:32:55,001
four or five difference in the distances.
As long as there's sort of, you know,

1666
01:32:55,001 --> 01:33:02,001
What was the ratio? Yeah, so what I did
in the thesis, I picked, picked the power

1667
01:33:02,001 --> 01:33:07,001
control. So it actually, if somebody's
five times closer than somebody you're

1668
01:33:07,001 --> 01:33:12,000
transmitting to, then the scheduling
constraint kicks in. And it was very rare.

1669
01:33:12,001 --> 01:33:16,000
As you can see, there are only a few
packets that we need to do about. As I turn

1670
01:33:16,000 --> 01:33:20,001
that on, here's the whole distribution
with what I call the respect neighbor

1671
01:33:20,001 --> 01:33:22,001
constraint to the scheduling.

1672
01:33:25,001 --> 01:33:32,000
And it was a tunable number, but I found
the number that just works. It doesn't,

1673
01:33:32,001 --> 01:33:37,000
it hardly changes anything. The distribution
doesn't really move anywhere. I

1674
01:33:37,000 --> 01:33:40,001
mean, at this scale, it looks the same.
You can't tell the difference. But then

1675
01:33:40,001 --> 01:33:44,000
when you zoom in on the tail and you
look at the worst case signal to noise

1676
01:33:44,000 --> 01:33:49,000
ratios, the worst one was around.
It cut off, which means all the packets were

1677
01:33:49,000 --> 01:33:53,000
delivered here, which means that if you had,
by my model for what a successful

1678
01:33:53,000 --> 01:33:57,001
reception is, if you had processing gain
here for that sort of factor of three

1679
01:33:57,001 --> 01:34:02,000
fudge factor, maybe five dB, by my model,
I can wave my hands and say, by my

1680
01:34:02,000 --> 01:34:08,001
model, no packets are dropped due to
collisions. Or due to interference, or

1681
01:34:08,001 --> 01:34:15,000
whatever.
And that's what I mean by collision free. So, I

1682
01:34:15,000 --> 01:34:19,000
also did simulations with a thousand stations.

1683
01:34:22,000 --> 01:34:29,000
Same idea. You've seen all these plots
before with a hundred, but here's the, I

1684
01:34:29,000 --> 01:34:34,001
have power control turned on already.
And you get a distribution that looks like

1685
01:34:34,001 --> 01:34:38,001
that. The tail of this distribution goes
a little further. And that's because

1686
01:34:38,001 --> 01:34:44,000
with a thousand stations, the closest,
the random guy who, you know, we wound up

1687
01:34:44,000 --> 01:34:47,000
with two guys sitting proportionally
closer to each other than

1688
01:34:47,000 --> 01:34:49,000
we did in the thing. And that'll, that'll work.

1689
01:34:49,000 --> 01:34:54,000
And that'll happen as you have more chances
for that to happen. So, so I'm going

1690
01:34:54,000 --> 01:34:58,000
to, again, zoom in on the lower left tail
of that. And you can see that there's

1691
01:34:58,000 --> 01:35:03,001
some fairly low signal to noise ratios.
And here's the minus 20 dB. And so

1692
01:35:03,001 --> 01:35:08,000
there's actually, we cut off at about
minus 17 before. And there's, we've

1693
01:35:08,000 --> 01:35:11,000
actually got quite a, quite a lot out there.
And I turn on the respect neighbor

1694
01:35:11,000 --> 01:35:14,001
constraint, which again, rarely comes
into play. There's a lot of mass in this

1695
01:35:14,001 --> 01:35:18,000
distribution to the right of here.
Even though we see a lot of stuff here.

1696
01:35:19,000 --> 01:35:22,001
And turn on the respect neighbor. What?

1697
01:35:26,000 --> 01:35:29,001
Yeah, that's right. And you turn on the
respect neighbor constraint here in the

1698
01:35:29,001 --> 01:35:36,000
thousand station case. And again,
it cuts off at about minus 17 dB. Now,

1699
01:35:37,001 --> 01:35:42,000
gosh, it's been a while. There was a
time when I convinced myself that as you

1700
01:35:42,000 --> 01:35:45,001
turn up the number of stations in the network,
and you have larger and larger

1701
01:35:45,001 --> 01:35:52,000
contributions from distance stations to
those sort of total noise, that this, as

1702
01:35:52,000 --> 01:35:57,001
this walks left, as the sort of whole
distribution walks left, you can, you can

1703
01:35:57,001 --> 01:36:03,001
cut it off tighter. So this actually
doesn't move much. This minus 17 dB. It

1704
01:36:03,001 --> 01:36:07,000
moves over more slowly than the whole
distribution moves over. So the respect

1705
01:36:07,000 --> 01:36:11,001
neighbor constraint is a fairly powerful thing.
And that's why I claim that all

1706
01:36:11,001 --> 01:36:14,000
packets are received successfully by this

1707
01:36:14,000 --> 01:36:16,001
model. What's the difference between these two?

1708
01:36:19,000 --> 01:36:24,001
I have a surprise slide in here that I
don't know what to steer for. I think I

1709
01:36:24,001 --> 01:36:29,001
have two duplicate slides. No, no,
they're the same. They're the same because

1710
01:36:29,001 --> 01:36:36,000
once the zoom in.
That's the one that's cut off. So,

1711
01:36:37,000 --> 01:36:43,000
where did Bob go? Back in the room now.
Your question earlier was, what about the

1712
01:36:43,000 --> 01:36:45,000
hotspot? The middle or not?

1713
01:36:45,000 --> 01:36:51,001
Let me find my thousand station picture. I

1714
01:36:51,001 --> 01:36:58,001
had a better picture earlier. Thousand

1715
01:36:58,001 --> 01:37:04,000
stations, just to think about for a second.
There are a thousand points in this

1716
01:37:04,000 --> 01:37:10,001
network. There are a million
stations to station paths. Because

1717
01:37:10,001 --> 01:37:14,000
there's a, or there's 450,000 or something.

1718
01:37:14,000 --> 01:37:19,001
It's one, it's roughly, oh, one way path.
There's a, if you include the path to

1719
01:37:19,001 --> 01:37:23,000
yourself, there's a million. Because
there's a thousand sources and a thousand

1720
01:37:23,000 --> 01:37:29,000
destinations. Subtract out the thousand
trivial ones. We're left with almost a

1721
01:37:29,000 --> 01:37:35,000
million. There's about a million paths in this.
And so what I did is I went

1722
01:37:35,000 --> 01:37:39,001
through, so there's a million start,
source the destination paths.

1723
01:37:40,001 --> 01:37:45,001
I went through and I traced every one of them.
And every time I came across the

1724
01:37:45,001 --> 01:37:51,000
station, I incremented a counter.
So now, for all thousand stations,

1725
01:37:52,001 --> 01:37:58,001
what I have is how many paths go through me.
So, and that's always an even

1726
01:37:58,001 --> 01:38:03,001
number. Because it's symmetric. That's

1727
01:38:03,001 --> 01:38:05,001
what I did.

1728
01:38:05,001 --> 01:38:11,000
And I've lost the, here it is.

1729
01:38:12,000 --> 01:38:17,001
Now, for my thousand stations,
what does the distribution look

1730
01:38:17,001 --> 01:38:19,000
like for the results of those numbers?

1731
01:38:25,001 --> 01:38:31,000
So, let's see. On this axis, I have number
of paths, which goes out to about 200

1732
01:38:31,000 --> 01:38:37,000
,000, or about 20%.
And there were a few cases over here.

1733
01:38:38,000 --> 01:38:42,000
Where there were, there were three stations
in this, in the middle of this. That

1734
01:38:42,000 --> 01:38:46,000
had over 20%
of the million paths going through them.

1735
01:38:49,000 --> 01:38:56,000
This is a hotspot in the routing.
And there were, you know, quite a few that

1736
01:38:56,000 --> 01:39:00,001
were over 100,000. Had more than 10%
of the paths through them. I actually went

1737
01:39:00,001 --> 01:39:04,000
through, and the thesis is a plot.
I don't have the foil for it, but there's a

1738
01:39:04,000 --> 01:39:09,000
plot that shows actually which stations had,
I think, more than 10%. So all the

1739
01:39:09,000 --> 01:39:14,001
stations that contributed to here,
I drew the plot. I drew this plot. But I left,

1740
01:39:14,001 --> 01:39:17,000
I didn't draw the square for the
station unless it was one of

1741
01:39:17,000 --> 01:39:19,000
those. So you can actually see where they are.

1742
01:39:21,000 --> 01:39:24,001
This is frustrating because this seems
to be like the bottleneck limiting sort

1743
01:39:24,001 --> 01:39:29,001
of, if you're doing end-to-end traffic in
the net. I think something can be done

1744
01:39:29,001 --> 01:39:36,001
to solve this. But I tried something
while I was doing the thesis. And

1745
01:39:36,001 --> 01:39:40,001
what I came up with was, well,
maybe I need to compute something other than

1746
01:39:40,001 --> 01:39:45,000
minimum energy routes. In fact, maybe what
we want to do is, is go twice as far.

1747
01:39:46,000 --> 01:39:48,000
Well, you could put a wormhole there, right?

1748
01:39:48,001 --> 01:39:53,000
Well, you could lay fiber across town,
absolutely. Or hop up to another frequency

1749
01:39:53,000 --> 01:40:00,000
and put a point to one at Microlink
across town. So what I did is I, the

1750
01:40:00,000 --> 01:40:03,000
one thing, the one experiment I had time
to do and write up and put in the thesis

1751
01:40:03,000 --> 01:40:07,001
was, I added a knob to the routing.
So that instead of minimum energy routing,

1752
01:40:07,001 --> 01:40:12,001
I'm doing minimize energy plus number,
some constant times number of hops.

1753
01:40:13,001 --> 01:40:18,000
And that constant is a knob I was able to turn.
And I turned that knob until the

1754
01:40:18,000 --> 01:40:20,001
average path links had dropped in half.
Meaning we're

1755
01:40:20,001 --> 01:40:22,000
going about twice as far in every hop.

1756
01:40:25,000 --> 01:40:30,001
This still has the property that a suffix
of a route is the route of that suffix.

1757
01:40:31,000 --> 01:40:33,000
So you can still have just a routing table and

1758
01:40:33,000 --> 01:40:35,000
everything. That took me well to think about.

1759
01:40:36,000 --> 01:40:40,001
And so I did this simulation. Now this is
the same simulation as before with the

1760
01:40:40,001 --> 01:40:43,000
Thousand Stations,
but it's plotted at the same scale as my next

1761
01:40:43,000 --> 01:40:45,000
slide so we can compare them.

1762
01:40:45,001 --> 01:40:48,001
This is the Thousand Station simulation
with power control and respect neighbor

1763
01:40:48,001 --> 01:40:53,001
turned on. This is the Thousand Station
simulation with power control, respect

1764
01:40:53,001 --> 01:40:58,001
neighbor,
and my modified routing that only goes,

1765
01:40:59,000 --> 01:41:00,001
that goes twice as far and a half.

1766
01:41:02,001 --> 01:41:06,000
I didn't print out the plot, but it also,
it reduced the sharing by about a

1767
01:41:06,000 --> 01:41:10,000
factor of two. And instead of having
the worst stations be 20% of the paths

1768
01:41:10,000 --> 01:41:13,000
through them, the worst stations are only 10%
of the paths through them.

1769
01:41:13,001 --> 01:41:18,001
So the difference is, is that this whole
distribution moved over about 5 dB.

1770
01:41:21,001 --> 01:41:26,001
So, or you'd have to reduce the,
increase the processing gain so you're lowering

1771
01:41:26,001 --> 01:41:30,000
the data rate that you could actually
send that and give them a fixed size.

1772
01:41:31,001 --> 01:41:35,000
So, so we have about a 5 dB difference in
signal noise ratio. Remember I said the

1773
01:41:35,000 --> 01:41:41,000
bit rate that you can send has a proportional
signal noise ratio. So you can, so

1774
01:41:41,000 --> 01:41:43,001
about a, so the way to actually think
about it is if you're going about twice as

1775
01:41:43,001 --> 01:41:46,001
far per hop,
that means you've turned your transmitter

1776
01:41:46,001 --> 01:41:48,001
up about a factor of four in power.

1777
01:41:49,000 --> 01:41:53,001
That'd be 6 dB.
So actually we should expect about a 6 dB

1778
01:41:53,001 --> 01:41:59,001
worsening in the signal to noise ratio.
And what we've done in the middle of the

1779
01:41:59,001 --> 01:42:03,000
network, so that means we can only
send that one fourth of the data

1780
01:42:03,000 --> 01:42:04,001
rate we would have been able
to in the other network.

1781
01:42:04,001 --> 01:42:07,001
So we've had a factor of four penalty
in signal to noise ratio. However, we

1782
01:42:07,001 --> 01:42:12,000
picked up a factor of two in sort of the,
the congestion

1783
01:42:12,000 --> 01:42:13,001
performance in the middle of the network.

1784
01:42:14,000 --> 01:42:18,001
Unfortunately it's in that loss of a
factor of two in total system performance,

1785
01:42:19,000 --> 01:42:21,000
assuming that what you're
concerned about is throughput.

1786
01:42:25,000 --> 01:42:28,001
So, shocks. It seems to me that it's important

1787
01:42:28,001 --> 01:42:30,000
to actually take the minimum energy route.

1788
01:42:32,001 --> 01:42:37,000
I tried to figure out, and never got anywhere,
and there's nothing in the thesis

1789
01:42:37,000 --> 01:42:40,001
about this, but I tried to figure out what,
how would you compute the optimum

1790
01:42:40,001 --> 01:42:42,001
routes given configuration.

1791
01:42:43,001 --> 01:42:47,000
I can't even figure out how to write down
the problem right so that people who

1792
01:42:47,000 --> 01:42:48,001
are really good at algorithms could help me.

1793
01:42:51,000 --> 01:42:57,001
It strikes me that what you need to do
is to apply this method to a certain

1794
01:42:57,001 --> 01:43:03,001
size network, certain number of units,
and then start

1795
01:43:03,001 --> 01:43:06,001
using a collapse factor on the unit.

1796
01:43:07,000 --> 01:43:12,000
Yes, absolutely. Have ways of getting past
the middle of the net. I'll still wave

1797
01:43:12,000 --> 01:43:14,001
my hands at the end though about
something I think would be interesting.

1798
01:43:15,000 --> 01:43:20,000
This led me to believe that minimum energy
routing is important, and maybe some

1799
01:43:20,000 --> 01:43:22,001
more work to do here.
But for now I believe in it.

1800
01:43:23,001 --> 01:43:25,001
Anyway, that's sort of the, this is not the

1801
01:43:25,001 --> 01:43:27,000
end of my talk,
this is sort of the end of my thesis.

1802
01:43:27,000 --> 01:43:31,000
Sort of the conclusion of my thesis is
that spread spectrum plus the multi-hop

1803
01:43:31,000 --> 01:43:34,001
routing and the scheduling method and
the minimum energy routing led to a

1804
01:43:34,001 --> 01:43:36,001
functional, scalable packet radio network.

1805
01:43:37,000 --> 01:43:39,001
And scale by scalable,
I mean millions of stations.

1806
01:43:42,001 --> 01:43:45,000
I still have, okay.

1807
01:43:49,001 --> 01:43:53,001
For a metro area, yeah.

1808
01:44:01,000 --> 01:44:01,001
Packets.

1809
01:44:05,001 --> 01:44:11,001
Can you give us a sense of the people on
the edges of the metropolitan area and

1810
01:44:11,001 --> 01:44:17,000
people towards the middle of the metropolitan
area, how much proportion they are

1811
01:44:17,000 --> 01:44:20,000
relaying other people and how
much they have originated?

1812
01:44:20,001 --> 01:44:23,001
Well, what you really care about is how
much performance do you get in the system

1813
01:44:23,001 --> 01:44:25,001
scales. And I'm going to do that.

1814
01:44:27,000 --> 01:44:29,001
Okay, so now I started thinking
about what we could build.

1815
01:44:30,001 --> 01:44:32,000
And I want to get my back
to my thousand station

1816
01:44:32,000 --> 01:44:36,000
picture because I'm going to wave my hands.

1817
01:44:38,000 --> 01:44:41,000
No, what you're asking is exactly the
right question. I have a way of addressing

1818
01:44:41,000 --> 01:44:44,001
that, which is still coming later in my talk.
Not there yet.

1819
01:44:46,000 --> 01:44:46,001
Good question.

1820
01:44:48,000 --> 01:44:52,001
Actually, the ratio, I'm not so much
concerned about how much traffic is my own

1821
01:44:52,001 --> 01:44:54,001
that I'm sending and how much am I relaying.

1822
01:44:55,000 --> 01:44:56,001
What I care about is how much do I get.

1823
01:45:01,000 --> 01:45:05,000
But if you're not happy when I get done
with the talk, ask your question again.

1824
01:45:08,000 --> 01:45:13,000
So I started thinking about what can we build.
It was originally this 47 CFR 15

1825
01:45:13,000 --> 01:45:17,001
.247 authorization that sort of inspired
me to go off and figure all this out.

1826
01:45:17,001 --> 01:45:20,000
Now I go try to fit this
system in that allocation.

1827
01:45:21,001 --> 01:45:25,000
And it's sort of frustrating.
The widest of those authorizations is

1828
01:45:25,000 --> 01:45:26,001
only 125 MHz wide.

1829
01:45:27,000 --> 01:45:30,000
And with 23DV processing gain,
the raw transmitter data rate is going to be

1830
01:45:30,000 --> 01:45:31,001
around 625 kilos per second.

1831
01:45:31,001 --> 01:45:35,001
A little bit of hand waving here,
but I'm not off by more than a factor of two.

1832
01:45:36,000 --> 01:45:40,000
So the maximum throughput to a single
neighbor is about 94 kilos per second.

1833
01:45:40,001 --> 01:45:44,000
And actually, there's a factor that you
didn't see, but this is basically what

1834
01:45:44,000 --> 01:45:48,000
fraction in that plot where I had
all the received duty cycles.

1835
01:45:49,000 --> 01:45:52,000
With another factor thrown in for the
actual fitting of the packet bin because

1836
01:45:52,000 --> 01:45:54,001
you can't always put full-size packets in.

1837
01:45:55,000 --> 01:46:00,000
Anyway, so you can only go to a single
neighbor at about 94 kilos per second.

1838
01:46:00,000 --> 01:46:02,001
So the way to think of
that is what you've got is

1839
01:46:02,001 --> 01:46:04,000
you've got a net that kind of looks like that.

1840
01:46:04,001 --> 01:46:06,001
And every one of those little edges is about a

1841
01:46:06,001 --> 01:46:08,001
94 kilos per second bi-directional pipe.

1842
01:46:09,000 --> 01:46:12,001
And now you can start thinking about
traffic in terms of nodes and edges.

1843
01:46:13,001 --> 01:46:17,000
That's kind of frustrating because
that's got to be shared by

1844
01:46:17,000 --> 01:46:19,000
200,000 paths through it.

1845
01:46:19,000 --> 01:46:21,001
And I don't know how many of those
actually have traffic on them.

1846
01:46:24,000 --> 01:46:27,000
So think, well, what could I do? Well,
maybe, you know, there's all sorts

1847
01:46:27,000 --> 01:46:28,001
of millimeter wave spectrum up there.

1848
01:46:28,001 --> 01:46:33,001
And if we wait a while, maybe we'll be
able to build stuff there sometime.

1849
01:46:35,000 --> 01:46:40,000
And so a two gigahertz wide system with
23U processing gain, you can have a raw

1850
01:46:40,000 --> 01:46:42,000
transmitter data of about
10 megabits per second.

1851
01:46:44,000 --> 01:46:46,001
Maximum throughput to a single neighbor
would be about 1.5 megabits per second.

1852
01:46:46,001 --> 01:46:49,000
So that's similar to a T1 leased line.

1853
01:46:49,000 --> 01:46:50,001
But still, you've got to share that.

1854
01:46:50,001 --> 01:46:52,001
And it's not clear to me
that this is interesting.

1855
01:46:53,000 --> 01:46:55,000
It doesn't seem like in a million-station

1856
01:46:55,000 --> 01:46:57,000
network, you're getting enough performance.

1857
01:46:57,000 --> 01:46:59,000
So you might have a rush in the body.

1858
01:47:00,000 --> 01:47:03,001
I thought I might be able to do that.
You don't do it on my direction.

1859
01:47:04,000 --> 01:47:09,001
Yeah, well, so what I didn't have time,
another thing I didn't have time to

1860
01:47:09,001 --> 01:47:12,000
explore in the thesis is what is
the effect of directional antennas?

1861
01:47:15,000 --> 01:47:20,000
And the way to think about this is,
it's just 12 dB of directional gain.

1862
01:47:20,000 --> 01:47:22,000
And that means 6 dB at each end.

1863
01:47:22,001 --> 01:47:24,000
This is directional gain over Omni.

1864
01:47:24,001 --> 01:47:26,000
And this is very interesting because...

1865
01:47:30,000 --> 01:47:35,001
So it's Don, right? Don's talk two days ago,
I was sitting here listening to it,

1866
01:47:35,001 --> 01:47:40,000
and I'm like, oh, he started with
directionality and then kind of mixed by, you

1867
01:47:40,000 --> 01:47:42,001
know, by the end you realize maybe
he wanted to add spread spectrum.

1868
01:47:42,001 --> 01:47:45,001
I started with spread spectrum and Omni,
and I got to the end and

1869
01:47:45,001 --> 01:47:47,000
think, well, maybe I really want directional.

1870
01:47:47,000 --> 01:47:53,000
And it's kind of like, I think we wind
up at the same place once we've really

1871
01:47:53,000 --> 01:47:55,000
thought everything through and thought
through the stuff we'd understand.

1872
01:47:56,000 --> 01:48:00,000
But really, I think all we need for
directional gain is about 12 dB.

1873
01:48:01,001 --> 01:48:04,001
And that's, I mean, 6 dB at the
transmitter and 6 dB at the receiver.

1874
01:48:05,000 --> 01:48:09,001
And this is 6 dB over Omni, meaning that
I've got a sort of a 6 dB main lobe

1875
01:48:09,001 --> 01:48:12,001
that, such that everything that's
not in that main lobe is down 6

1876
01:48:12,001 --> 01:48:14,000
dB from what I've got in that main lobe.

1877
01:48:14,000 --> 01:48:17,001
And you do that at the transmitter and
receiver, and you can add those together.

1878
01:48:20,000 --> 01:48:25,001
Exactly. So if you could do hard sector,
hard quadrants as sectors, you'd only

1879
01:48:25,001 --> 01:48:28,000
have to have, like, you know,
look in four different directions.

1880
01:48:33,001 --> 01:48:34,001
Exactly what I'm thinking.

1881
01:48:35,000 --> 01:48:38,000
Now, in fact,
you probably wouldn't be able to produce

1882
01:48:38,000 --> 01:48:40,000
hard sectors, but you probably built six.

1883
01:48:40,000 --> 01:48:42,000
And each of them would have
sort of a 6 dB main lobe.

1884
01:48:42,000 --> 01:48:45,001
And maybe you could actually get
maybe another dB out of it with

1885
01:48:45,001 --> 01:48:48,000
6, or maybe a half a dB or something.

1886
01:48:48,001 --> 01:48:53,001
But anyway, now, maybe we could get
through with maybe 10 or maybe 12 dB of

1887
01:48:53,001 --> 01:48:55,001
spread spectrum processing gain
with the directional gain.

1888
01:48:55,001 --> 01:48:58,000
You'd absolutely have to wrap this
up with good error correcting codes

1889
01:48:58,000 --> 01:49:00,000
with this little processing gain.

1890
01:49:00,000 --> 01:49:01,000
But I think it's doable.

1891
01:49:02,000 --> 01:49:06,000
And then a 2 GHz wide system could have a
raw throughput when your transmitter is

1892
01:49:06,000 --> 01:49:08,000
turned on of about 200 megabits per second.

1893
01:49:08,000 --> 01:49:12,000
And so the maximum throughput to a single
main lobe would be, actually, more

1894
01:49:12,000 --> 01:49:14,000
recently,
I've been saying about 25 megabits per second.

1895
01:49:14,000 --> 01:49:17,000
These are kind of hand waving numbers
because they, you know, it all depends on

1896
01:49:17,000 --> 01:49:19,000
what the traffic is and a
lot of things like that.

1897
01:49:19,000 --> 01:49:22,000
I like to think of,
when I made the slide, this slide

1898
01:49:22,000 --> 01:49:23,001
is more than three years old.

1899
01:49:25,000 --> 01:49:26,001
These days I'm thinking 25 megabits per second.

1900
01:49:27,000 --> 01:49:33,000
So now, let's think about what the
capacity of a system would be.

1901
01:49:33,000 --> 01:49:35,001
Now, what I've got on this slide
is only a thousand stations.

1902
01:49:35,001 --> 01:49:38,000
But imagine this is a million stations.

1903
01:49:38,000 --> 01:49:39,001
I'm not actually going to
point at any one station.

1904
01:49:39,001 --> 01:49:41,000
I'm just going to wave at
the whole network here.

1905
01:49:41,001 --> 01:49:47,001
If this is a million stations and it's
like a metropolitan area, the number,

1906
01:49:48,001 --> 01:49:51,000
the bottleneck is kind of
getting across the middle.

1907
01:49:51,000 --> 01:49:54,001
Now, I haven't figured out how to do the
routing to spread all the traffic out.

1908
01:49:54,001 --> 01:49:59,000
But imagine, but what is the total capacity
from the left side to the right side?

1909
01:49:59,001 --> 01:50:04,001
Well, cut it down the middle and ask
how many of these 25 megabits per

1910
01:50:04,001 --> 01:50:06,000
second bidirectional links you have cut.

1911
01:50:06,001 --> 01:50:09,001
I've thought about it and my intuition is
it ought to be about the square root of

1912
01:50:09,001 --> 01:50:11,001
the number of stations or proportional to that.

1913
01:50:11,001 --> 01:50:14,000
But it really ought to be,
wave my hands a little bit.

1914
01:50:14,001 --> 01:50:18,001
I actually took a bunch of these plots,
drew a line across them and counted, and

1915
01:50:18,001 --> 01:50:21,000
always came out to be right around the
square root of the number of stations.

1916
01:50:22,000 --> 01:50:23,001
So I believe that number now.

1917
01:50:24,000 --> 01:50:25,000
It matches intuition.

1918
01:50:26,000 --> 01:50:28,001
So if you've got a million
stations and you cut it down the

1919
01:50:28,001 --> 01:50:30,000
middle, you cut about a thousand links.

1920
01:50:32,001 --> 01:50:36,000
If each of those thousand links is a 25
megabits per second bidirectional pipe,

1921
01:50:36,001 --> 01:50:40,000
what you have cut is 25
gigabits per second of capacity

1922
01:50:40,000 --> 01:50:42,000
from the left half to the
right half of the system.

1923
01:50:44,000 --> 01:50:45,000
So that's the bottleneck.

1924
01:50:45,000 --> 01:50:48,000
We've only got 25 gigabits per second
capacity in the middle of the net.

1925
01:50:48,000 --> 01:50:50,000
We've got half a million users on each side.

1926
01:50:53,000 --> 01:50:54,001
How should we think about that?

1927
01:50:54,001 --> 01:51:00,000
Well, the way to think about that is take
the 25 gigabits per second and go ahead

1928
01:51:00,000 --> 01:51:02,001
and do the division divided by the 500,000.

1929
01:51:02,001 --> 01:51:05,000
You come out with a number which
is in units of bits per second.

1930
01:51:05,001 --> 01:51:07,000
Actually, I've got this on the next slide.

1931
01:51:08,001 --> 01:51:11,001
All these numbers I was just saying in
the past couple minutes. 200 megabits per

1932
01:51:11,001 --> 01:51:15,001
second, 25 megabits per second bidirectional
links, shared 25 gigabits per second

1933
01:51:15,001 --> 01:51:18,000
capacity across the middle from the left half

1934
01:51:18,000 --> 01:51:19,001
to the right half, or top to bottom.

1935
01:51:19,001 --> 01:51:20,001
It doesn't matter where you slice it.

1936
01:51:20,001 --> 01:51:22,000
You've got a bottleneck from there to there.

1937
01:51:22,001 --> 01:51:27,001
So the per user, if you divide,
do the division, 25 gigabits per second divided

1938
01:51:27,001 --> 01:51:30,000
by half a million,
you get 50 kilobits per second.

1939
01:51:33,000 --> 01:51:34,001
Sort of per user data rate.

1940
01:51:34,001 --> 01:51:39,000
Now, how to understand that is to think about,
well, what's the best connected

1941
01:51:39,000 --> 01:51:41,000
unit, you know,
if you wanted to really be on the internet.

1942
01:51:42,000 --> 01:51:42,001
Where do you go?

1943
01:51:42,001 --> 01:51:45,001
Well, you go to some major university
that's got like a big pipe.

1944
01:51:47,000 --> 01:51:49,001
And actually, this number is changing.

1945
01:51:49,001 --> 01:51:53,001
But if you take those big pipes,
when I first did this calculation three years

1946
01:51:53,001 --> 01:51:59,001
ago, the big universities were typically
collected over maybe a DS3, which is

1947
01:51:59,001 --> 01:52:02,000
about 50 megabits per second,
45 megabits per second.

1948
01:52:02,001 --> 01:52:04,001
You do the division on a university
campus and you start coming out with

1949
01:52:04,001 --> 01:52:06,001
numbers that are about 5 kilobits per second.

1950
01:52:07,000 --> 01:52:09,000
You take that pipe size,
divide it by the number of people on

1951
01:52:09,000 --> 01:52:11,001
campus, about 5 kilobits per second.

1952
01:52:11,001 --> 01:52:18,000
So this system seems to have 10 times
the capacity of a major university

1953
01:52:18,000 --> 01:52:20,001
in the mid 1990s, had to the internet.

1954
01:52:22,000 --> 01:52:25,001
And that's exciting because that means
this sort of can provide capacity

1955
01:52:25,001 --> 01:52:27,001
in a metropolitan area.

1956
01:52:27,001 --> 01:52:30,000
If you could build the 2 gigahertz wide system,
and I don't think we have the

1957
01:52:30,000 --> 01:52:31,001
semiconductor to do that yet, but...

1958
01:52:32,000 --> 01:52:34,001
Just like all students are
and always on the computer.

1959
01:52:35,000 --> 01:52:37,001
Yeah, yeah.
So then, so I mean, yeah, you're sharing this.

1960
01:52:37,001 --> 01:52:40,001
And so it's just like the telephone system.
I think they typically engineer for

1961
01:52:40,001 --> 01:52:43,000
only 10 people to pick up their phone
and make a call at any given time.

1962
01:52:44,000 --> 01:52:44,001
Question here.

1963
01:52:44,001 --> 01:52:51,000
In the starting supposition that everybody
has equal traffic on the network...

1964
01:52:51,000 --> 01:52:54,001
I did not... ...with a number of messages.

1965
01:52:57,000 --> 01:53:00,000
What happens if in reality,
like there is on the internet,

1966
01:53:00,000 --> 01:53:02,000
where you have some major site?

1967
01:53:03,000 --> 01:53:07,001
I don't know. I remember, you know,
the last things in my thesis is

1968
01:53:07,001 --> 01:53:09,000
like, until we have real user generating

1969
01:53:09,000 --> 01:53:10,001
traffic, we have no idea what
the traffic is going to be.

1970
01:53:11,000 --> 01:53:12,001
But yeah,
there'll probably be some concentration.

1971
01:53:13,001 --> 01:53:18,000
But this is an interesting system.
It's really all the only argument I'm trying

1972
01:53:18,000 --> 01:53:20,000
to make here.
It's an interesting amount of capacity.

1973
01:53:20,001 --> 01:53:24,001
The other issue for this, again,
thinking about this, million stations.

1974
01:53:25,001 --> 01:53:28,001
If this is a million stations and you
want to get from one side to the other,

1975
01:53:28,001 --> 01:53:31,001
well, you got it. The number of hops
you got to go is about the, again,

1976
01:53:31,001 --> 01:53:33,000
the square root of the number of stations.

1977
01:53:34,000 --> 01:53:36,001
So to get across the metropolitan area of
the million stations, you might have to

1978
01:53:36,001 --> 01:53:40,000
go a thousand hops to get from one
side to the other, all the way across.

1979
01:53:42,000 --> 01:53:48,000
And in a 2 GHz wide system with 200 mbps
raw transmitter data rate, 500 bits per

1980
01:53:48,000 --> 01:53:52,000
packet, the packet transmission time is only 2.
5 microseconds.

1981
01:53:54,001 --> 01:53:59,001
Now, but you've got to wait in
pseudo-random schedules until you can send

1982
01:53:59,001 --> 01:54:01,000
it. You've actually got to wait
some number of slot times.

1983
01:54:01,000 --> 01:54:04,000
I'm thinking 10 microsecond slots here.
And you've got to wait about, on

1984
01:54:04,000 --> 01:54:06,000
average, four, maybe five slots.

1985
01:54:06,001 --> 01:54:09,001
So it takes about a scheduling delay
of about 50 microseconds per hop.

1986
01:54:11,000 --> 01:54:14,000
And if you're going to go a thousand hops,
that's 50 milliseconds. One

1987
01:54:14,000 --> 01:54:16,000
way delay across the metropolitan area.

1988
01:54:16,001 --> 01:54:18,001
So you could get everybody
talking to everybody.

1989
01:54:18,001 --> 01:54:25,001
So this, the conclusion between this capacity,
which is 50 kbps, that's more

1990
01:54:25,001 --> 01:54:27,001
trunk, 10 times more capacity than typical

1991
01:54:27,001 --> 01:54:29,000
well-conversed universities on the Internet.

1992
01:54:29,001 --> 01:54:34,000
It's also way more capacity than a telephone
company in a large metropolitan area

1993
01:54:34,000 --> 01:54:36,001
would have to have in terms
of trunking to support

1994
01:54:36,001 --> 01:54:38,000
all the voice calls in the metropolitan area.

1995
01:54:38,001 --> 01:54:42,001
So this is a system,
the delay, you can do voice

1996
01:54:42,001 --> 01:54:44,001
with 100 millisecond round trip time.

1997
01:54:45,000 --> 01:54:48,000
So this system has more capacity
in the metropolitan area telephone

1998
01:54:48,000 --> 01:54:50,000
system. Voice.

1999
01:54:51,001 --> 01:54:54,000
And that's really neat.
We could cancel our phone service

2000
01:54:54,000 --> 01:54:55,001
and just talk over this.

2001
01:54:56,000 --> 01:55:01,000
And long distance telephone companies
could provide, you know, packet voice long

2002
01:55:01,000 --> 01:55:04,000
distance, like Internet long
distance or something, I don't

2003
01:55:04,000 --> 01:55:05,001
know, across this thing.

2004
01:55:07,000 --> 01:55:09,000
They could come into an area and just
put up a few points of presence.

2005
01:55:09,000 --> 01:55:11,001
Or maybe if they wanted to
provide better service with less

2006
01:55:11,001 --> 01:55:13,000
delay, they'd have to get closer to you.

2007
01:55:13,000 --> 01:55:15,001
And now you'd have a choice as to whether
you wanted to go all the way across

2008
01:55:15,001 --> 01:55:20,001
town with a cheap guy who's only got one
point of presence or the quality long

2009
01:55:20,001 --> 01:55:23,000
distance provider that's actually gone
in and put 100 things so they can

2010
01:55:23,000 --> 01:55:25,000
get, maybe even give you a higher capacity.

2011
01:55:25,000 --> 01:55:27,001
Now, how soon? I don't know.

2012
01:55:29,001 --> 01:55:33,000
When I finished the thesis, I said that
this will be feasible in 7 to 10 years.

2013
01:55:33,000 --> 01:55:36,001
You could actually, by feasible I
mean you could actually do this in

2014
01:55:36,001 --> 01:55:38,000
affordable consumer electronics.

2015
01:55:40,001 --> 01:55:45,001
And that's, you know, how soon can we
do a 2 GHz chipping rate in a receiver

2016
01:55:45,001 --> 01:55:48,001
that's doing error correcting
codes and everything in

2017
01:55:48,001 --> 01:55:50,001
BLSI. It'd be full cost in BLSI, of course.

2018
01:55:51,001 --> 01:55:54,001
So here we are three years later and I
think my prediction's about right, which

2019
01:55:54,001 --> 01:55:59,000
means subtract 3 from 7 and you get 4,
subtract 3 from 10. So it's 4 to 7 years.

2020
01:55:59,001 --> 01:56:01,001
I think this could be affordable.

2021
01:56:03,001 --> 01:56:08,000
And actually this is more my research talk.
This is like things I want to

2022
01:56:08,000 --> 01:56:10,001
understand.
What does real propagation look like?

2023
01:56:11,000 --> 01:56:14,000
I want to go out and measure impulse
responses for 2 GHz wide signals.

2024
01:56:17,000 --> 01:56:18,000
[...] guess of one the other.

2025
01:56:18,001 --> 01:56:22,000
Yeah. Does directionality really
work the way I think it does?

2026
01:56:22,000 --> 01:56:25,001
I've made this hand waving argument about
this directional gain, but I haven't

2027
01:56:25,001 --> 01:56:28,001
done any simulations there.
I've just tried to wrap my brain around it.

2028
01:56:28,001 --> 01:56:32,000
I really need to go off and actually come
up with the more robust argument that

2029
01:56:32,000 --> 01:56:35,001
could actually be written down about
using directionality as a trade

2030
01:56:35,001 --> 01:56:37,000
off for spread spectrum processing gain.

2031
01:56:37,001 --> 01:56:43,001
And maybe it's not 1 to 1. Maybe 12 dB
of directional gain total only gives you

2032
01:56:43,001 --> 01:56:47,001
about the ability to throw away,
to reduce your spread spectrum

2033
01:56:47,001 --> 01:56:49,000
processing gain by maybe 10 or 8 dB.

2034
01:56:50,000 --> 01:56:51,001
But it's certainly still worth it.

2035
01:56:52,000 --> 01:56:55,000
There's all these sorts of real networking
issues. And actually I was in graduate

2036
01:56:55,000 --> 01:56:58,001
school in a networking research group,
not a radio research group.

2037
01:56:59,000 --> 01:57:00,001
And so actually I went off and tackled the

2038
01:57:00,001 --> 01:57:02,000
problems that I thought were hard first.

2039
01:57:02,001 --> 01:57:05,001
And I was on the radio stuff
and the scalable radio system.

2040
01:57:06,000 --> 01:57:10,000
But there's all sorts of hard networking
problems about a completely ad hoc

2041
01:57:10,000 --> 01:57:15,001
network about how to actually,
how does the station get a name like by name I

2042
01:57:15,001 --> 01:57:17,000
need a unique ID, a pile of bets.

2043
01:57:18,000 --> 01:57:21,000
And then how do you do
routing across this system?

2044
01:57:21,000 --> 01:57:24,000
How do you actually compute the routes?
I mean you want something that produces

2045
01:57:24,000 --> 01:57:26,000
something close to minimum energy routing.

2046
01:57:26,001 --> 01:57:28,001
But how do you actually compute
the routes and how do you find the

2047
01:57:28,001 --> 01:57:30,000
path, how do you find which way to swing.

2048
01:57:31,000 --> 01:57:34,000
And then if somebody, if people in the
metropolitan area are offering services

2049
01:57:34,000 --> 01:57:36,001
available, how do they advertise those?

2050
01:57:37,000 --> 01:57:41,001
How do they even publicize or spread the
information that, hey, I'm not only just

2051
01:57:41,001 --> 01:57:45,000
going to address here, but if you actually
contact me I can provide some service.

2052
01:57:46,001 --> 01:57:49,000
And maybe the service you want to
get is an access to the Internet.

2053
01:57:49,001 --> 01:57:53,001
Maybe across this system somebody's offering
a service which is global worldwide

2054
01:57:53,001 --> 01:57:57,000
Internet connectivity where they'll
hand you an address which is globally

2055
01:57:57,000 --> 01:57:59,001
routable, maybe through some
protocol and now you can route.

2056
01:58:00,001 --> 01:58:03,001
Now IPv6 by the way is going to be
all automatic address configuration,

2057
01:58:03,001 --> 01:58:05,001
even all the way down through
the tree and the routers.

2058
01:58:06,000 --> 01:58:10,000
And so I imagine there'll be a global
Internet with IPv6 and that sort of thing.

2059
01:58:11,000 --> 01:58:14,000
How to fund further research and development?
I don't know.

2060
01:58:15,000 --> 01:58:16,001
I'm burning my savings account at the moment.

2061
01:58:17,001 --> 01:58:20,000
And then how can this be brought to the market?
One of these things, this thing

2062
01:58:20,000 --> 01:58:22,001
would be really cool if everybody had one,
but how do you sell the first

2063
01:58:22,001 --> 01:58:24,000
hundred in the metropolitan area?

2064
01:58:24,001 --> 01:58:26,001
Maybe you have to give them away.
Is that enough?

2065
01:58:27,000 --> 01:58:28,001
Is it interesting enough? I don't know.

2066
01:58:30,000 --> 01:58:35,001
Now the political talk, actually this is
a historical talk. What I think is going

2067
01:58:35,001 --> 01:58:40,001
on here, imagine in 1975,
think about the computer industry, the computer

2068
01:58:40,001 --> 01:58:45,000
industry, I think from about early 1960s,
certainly at IBM when they started

2069
01:58:45,000 --> 01:58:49,001
doing the plans for the IBM 360,
they understood that, hey, we're creating

2070
01:58:49,001 --> 01:58:51,000
a computer revolution, isn't this great?

2071
01:58:52,000 --> 01:58:56,000
And the world of the future is the
world after we're going to have a

2072
01:58:56,000 --> 01:58:58,000
computer and information processing revolution.

2073
01:58:58,001 --> 01:59:01,001
And everybody's going to be using computers.
But if you look at all the research

2074
01:59:01,001 --> 01:59:06,000
work and all the development work in the
late 60s, the vision was that everybody

2075
01:59:06,000 --> 01:59:08,001
in the future is going to be using computers.
And computers are big things that

2076
01:59:08,001 --> 01:59:11,000
fill machine rooms that need air
conditioning and use lots of power and need

2077
01:59:11,000 --> 01:59:12,001
tending and people got to
mount tapes and everything.

2078
01:59:13,000 --> 01:59:16,000
And so computers are going to be
centralized things. And then what they were

2079
01:59:16,000 --> 01:59:18,000
going to do is they were going to figure
out how to make computers reliable

2080
01:59:18,000 --> 01:59:22,000
enough and run time sharing on them
and offer computers as a utility

2081
01:59:22,000 --> 01:59:24,000
and you'd have your computer jack in the wall.

2082
01:59:26,001 --> 01:59:30,000
And the computer industry was a blooming
industry. There were companies that were

2083
01:59:30,000 --> 01:59:34,001
manufacturing computers and selling computers,
sort of as equipment providers,

2084
01:59:34,001 --> 01:59:40,000
and there were the companies that bought
computers and sold time on them for

2085
01:59:40,000 --> 01:59:42,001
people to come and run their payroll,
for people to come and do their scientific

2086
01:59:42,001 --> 01:59:45,000
calculations.
And the big companies, well, they could

2087
01:59:45,000 --> 01:59:46,001
afford to finance it internally themselves.

2088
01:59:46,001 --> 01:59:50,001
But for the vision of the future, at least,
is how everybody's going to be able

2089
01:59:50,001 --> 01:59:54,000
to get affordable computing is
that it was going to be offered

2090
01:59:54,000 --> 01:59:55,001
as a service that you paid for.

2091
01:59:56,000 --> 01:59:59,000
You had a monthly bill.
It was going to be your computer

2092
01:59:59,000 --> 02:00:00,001
bill. It was going to be utility on the wall.

2093
02:00:02,000 --> 02:00:05,001
It's really neat to read all the visionary
papers from the 60s because the future

2094
02:00:05,001 --> 02:00:09,001
is a world where everybody has access to
a computer, but it's off somewhere else.

2095
02:00:10,001 --> 02:00:17,000
And what happened is, in 1975,
people started proposing that maybe, actually,

2096
02:00:18,000 --> 02:00:21,001
you build your computer or you go to the
store and you buy your computer, you

2097
02:00:21,001 --> 02:00:26,001
write a check for it, or actually these
days it's plastic money, but you take

2098
02:00:26,001 --> 02:00:28,000
the computer home. You can afford it.

2099
02:00:29,000 --> 02:00:34,001
And you carry it home and use it.
And, of course, in 1975, that's a ridiculous

2100
02:00:34,001 --> 02:00:37,001
idea. You can't afford a real computer.

2101
02:00:38,000 --> 02:00:41,000
A computer is just a toy.
You can't do anything useful on it. In

2102
02:00:41,000 --> 02:00:42,001
fact, they were just toys at the beginning.

2103
02:00:42,001 --> 02:00:47,001
But, actually, there's a wonderful
book by Clayton Christensen called The

2104
02:00:47,001 --> 02:00:54,001
Innovator's Dilemma,
and he talks about how disruptive technology

2105
02:00:54,001 --> 02:01:00,000
displaces established industries, like sort

2106
02:01:00,000 --> 02:01:02,000
of stealth coming from below.

2107
02:01:02,000 --> 02:01:03,001
And the PC was like this.

2108
02:01:06,000 --> 02:01:09,001
And, anyway,
so here's my sort of grand statement. It's like

2109
02:01:09,001 --> 02:01:11,001
you could own your own
telecommunications network.

2110
02:01:12,000 --> 02:01:16,000
My vision is that the box that goes
on the house is something that you

2111
02:01:16,000 --> 02:01:18,001
go to the store and you buy one.

2112
02:01:18,001 --> 02:01:20,001
You take it home when you put it in your house,
and there's no monthly bill.

2113
02:01:21,000 --> 02:01:22,001
Well,
there's no monthly bill for long distance.

2114
02:01:23,000 --> 02:01:24,001
There's no monthly bill
for the metropolitan area.

2115
02:01:25,000 --> 02:01:27,001
And the way to think about that is that

2116
02:01:27,001 --> 02:01:31,000
what is, you know,
your bill for the long distance.

2117
02:01:31,000 --> 02:01:35,000
Most of what, even what you pay for a long
distance call today goes to the phone

2118
02:01:35,000 --> 02:01:37,001
company on your end and the
phone company at the far end.

2119
02:01:38,000 --> 02:01:40,001
And if you're paying nine cents a minute,
it's like four cents a minute to the

2120
02:01:40,001 --> 02:01:42,000
far end, four cents a minute to your end.

2121
02:01:42,001 --> 02:01:46,000
And the long distance company gets to
keep about a penny per minute to pay for

2122
02:01:46,000 --> 02:01:50,001
its infrastructure investment and all
of its operating costs and profits.

2123
02:01:51,001 --> 02:01:55,001
Long distance is actually very cheap,
except that they have to find a way to get

2124
02:01:55,001 --> 02:01:57,001
to you, the access.

2125
02:01:57,001 --> 02:01:59,001
Some people say the last model problem. It's

2126
02:01:59,001 --> 02:02:01,000
really the metropolitan area problem.

2127
02:02:02,000 --> 02:02:06,001
And I think that the right way to do this
is to, you know, own most of all your

2128
02:02:06,001 --> 02:02:11,000
telecommunications and perhaps global
worldwide internet connectivity for sort of

2129
02:02:11,000 --> 02:02:12,001
big packet data sort of stuff.

2130
02:02:12,001 --> 02:02:17,000
Maybe you can pay 60 bucks per year flat
rate for sort of unlimited service that

2131
02:02:17,000 --> 02:02:19,000
you can get across one of these things.

2132
02:02:19,001 --> 02:02:23,000
And maybe long distance calls, because
you actually care about the delay, maybe

2133
02:02:23,000 --> 02:02:26,001
that's something that would require that
you actually pay on a per call basis,

2134
02:02:26,001 --> 02:02:29,001
that we could get the, you know,
we could get the, maybe we could get the long

2135
02:02:29,001 --> 02:02:32,000
distance rates down to maybe
one or two cents a minute.

2136
02:02:32,000 --> 02:02:35,001
If the access on both ends was
across a network like this.

2137
02:02:36,000 --> 02:02:37,000
[...]

2138
02:02:38,001 --> 02:02:42,000
What? ...how they care about the long
distance companies plugging into...

2139
02:02:42,000 --> 02:02:43,000
This system?

2140
02:02:43,001 --> 02:02:47,000
Oh, there's a huge political problem there,
but fortunately, it's only a small

2141
02:02:47,000 --> 02:02:49,001
fraction of the world's
population that lives under

2142
02:02:49,001 --> 02:02:52,000
phone companies that are regulated by the FCC.

2143
02:02:54,001 --> 02:02:57,001
It's only, it's the five billion population
in the world, and there's only about

2144
02:02:57,001 --> 02:03:00,000
a little over a quarter billion in the US.

2145
02:03:00,001 --> 02:03:01,001
So there's a factor of 20.

2146
02:03:01,001 --> 02:03:04,001
So about, what's the five percent of
the world's population is in the US.

2147
02:03:06,000 --> 02:03:06,001
So...

2148
02:03:07,001 --> 02:03:11,000
So even just in the US, this sort of
idea of monthly bills versus no monthly

2149
02:03:11,000 --> 02:03:14,001
bills, one of the things that a lot of
people are talking about today, there's a

2150
02:03:14,001 --> 02:03:16,000
great feeling by a lot of people that we are

2151
02:03:16,000 --> 02:03:18,000
upon a telecommunications revolution.

2152
02:03:18,000 --> 02:03:20,000
The internet has just happened in a big way,

2153
02:03:20,001 --> 02:03:22,000
and there's all this, you know, mobile stuff.

2154
02:03:22,000 --> 02:03:23,001
What I've talked about is a fixed network.

2155
02:03:24,000 --> 02:03:27,001
I don't fully understand mobility yet
and what should be done about that.

2156
02:03:28,000 --> 02:03:29,001
Deploy this...

2157
02:03:29,001 --> 02:03:30,001
Just changing the subject for a moment.

2158
02:03:30,001 --> 02:03:35,001
A fixed network like this sure would be
a neat base from which maybe all these

2159
02:03:35,001 --> 02:03:39,001
boxes also have a little micro cell thing
from which you can then get in on some

2160
02:03:39,001 --> 02:03:42,001
other partition device up to
one of these to go across town.

2161
02:03:44,000 --> 02:03:44,001
But anyway...

2162
02:03:45,001 --> 02:03:48,000
I've run around and said,
no monthly bills, no monthly bills.

2163
02:03:48,001 --> 02:03:51,001
Actually, David Hughes is running
around and saying, free communication.

2164
02:03:52,001 --> 02:03:54,000
You know, this part 15 devices have it.

2165
02:03:54,001 --> 02:03:55,001
Actually, free is not the right word.

2166
02:03:55,001 --> 02:03:56,001
It's user financed.

2167
02:03:57,000 --> 02:04:02,000
The user paid for it all up front,
just like the PC is not free computing.

2168
02:04:02,001 --> 02:04:07,000
It is free computing compared to maybe
the point of view of a computer utility.

2169
02:04:07,000 --> 02:04:09,000
But you paid for it all up front.

2170
02:04:09,000 --> 02:04:10,000
It's user...

2171
02:04:10,001 --> 02:04:12,000
Yeah, it's user financed.

2172
02:04:12,001 --> 02:04:13,001
I guess you do have to pay the power bill.

2173
02:04:13,001 --> 02:04:18,000
Maybe you can pedal your bicycle or solar
panels, as we heard about last night.

2174
02:04:18,001 --> 02:04:21,000
But I actually think this version of,
you know, for

2175
02:04:21,000 --> 02:04:24,000
all of us, you know, for society...

2176
02:04:24,000 --> 02:04:26,000
I don't know, I'm rambling on here.

2177
02:04:26,001 --> 02:04:30,000
But I think the idea of no monthly
bills is a very important one because

2178
02:04:30,000 --> 02:04:32,001
telephone penetration in the
United States is somewhere

2179
02:04:32,001 --> 02:04:34,000
between 90 and 95 percent.

2180
02:04:34,001 --> 02:04:36,001
And the people who study this thing
get into big arguments about whether

2181
02:04:36,001 --> 02:04:38,000
it's 90 or 95 percent.

2182
02:04:39,001 --> 02:04:40,000
It's stunningly low.

2183
02:04:40,000 --> 02:04:41,000
I thought I'd already had a phone.

2184
02:04:41,001 --> 02:04:44,001
But actually, quite a few, you know,
maybe one in 20 or one in 10 households

2185
02:04:44,001 --> 02:04:46,000
in the U.S. do not have a working telephone.

2186
02:04:48,000 --> 02:04:49,000
And what's the problem?

2187
02:04:49,001 --> 02:04:54,000
Well, the standard mantra among those
who worry about this problem is

2188
02:04:54,000 --> 02:04:55,001
that it's the problem of affordability.

2189
02:04:56,001 --> 02:04:57,001
But it's not.

2190
02:04:57,001 --> 02:04:59,000
It's not the affordability.

2191
02:04:59,001 --> 02:05:02,000
It's that in order to have a telephone,
you have to have good credit.

2192
02:05:03,000 --> 02:05:04,000
Because a phone, you can run up a bill.

2193
02:05:04,000 --> 02:05:06,001
Or if you fail to pay, you know,
before they go in trouble putting in

2194
02:05:06,001 --> 02:05:08,001
service, then they'll actually want to
know they're going to get their money.

2195
02:05:08,001 --> 02:05:14,000
So it's actually easier to afford for
end users, for those who aren't terribly

2196
02:05:14,000 --> 02:05:19,000
wealthy or not the most wealthy in society,
it's easier for them to figure out

2197
02:05:19,000 --> 02:05:23,001
how to save up their money and pay to buy
the PC or buy the television than it is

2198
02:05:23,001 --> 02:05:28,000
to figure out how for them to reliably
enough pay a monthly bill for a service.

2199
02:05:28,001 --> 02:05:33,000
And so this sort of system, instead of
the cross subsidy thing, which is what

2200
02:05:33,000 --> 02:05:38,000
we've got now with the whole telephone,
you can still get a phone light, you

2201
02:05:38,000 --> 02:05:40,001
know, in some states you can still,
you know, if you

2202
02:05:40,001 --> 02:05:42,001
qualify for certain federal assistance.

2203
02:05:43,000 --> 02:05:48,000
I think this is actually a better way
to provide communication to everybody.

2204
02:05:48,001 --> 02:05:51,001
In fact, you know, I think back in the
70s there were probably people who

2205
02:05:51,001 --> 02:05:53,000
had CB radios who didn't have telephones.

2206
02:05:54,000 --> 02:05:55,001
And we have a question here.

2207
02:05:55,001 --> 02:05:57,000
That's television penetration.

2208
02:05:58,001 --> 02:06:02,000
Yeah, more people have,
that was the second point on my slide.

2209
02:06:02,000 --> 02:06:07,000
Apparently, television penetration in
the United States is like 99.5 or 99.8

2210
02:06:07,000 --> 02:06:09,001
percent of households have a television,
working television.

2211
02:06:10,001 --> 02:06:13,000
So, and there's no monthly bill
associated with a television.

2212
02:06:14,000 --> 02:06:15,000
You just go to the store and you

2213
02:06:15,000 --> 02:06:16,001
buy it. Well, you know,
they're paying for it with that.

2214
02:06:17,000 --> 02:06:20,001
They're paying for the, yeah, that's true.
Now England is different.

2215
02:06:21,000 --> 02:06:26,001
So, what about spectrum? Actually,
I believe 59 to 64 gigahertz is

2216
02:06:26,001 --> 02:06:28,001
already sort of open.

2217
02:06:28,001 --> 02:06:31,001
And maybe this whole system could be built
there. The problem there is you get a

2218
02:06:31,001 --> 02:06:35,001
15 dB per kilometer attenuation of a
free space loss because of an oxygen.

2219
02:06:35,001 --> 02:06:41,000
There's a quantum resonance. That would
be an advantage except it reduces the

2220
02:06:41,000 --> 02:06:43,001
maximum range of when you
start with two stations.

2221
02:06:43,001 --> 02:06:46,001
And so, actually, I don't think it,
it might not be the right place to start.

2222
02:06:47,000 --> 02:06:49,000
And I'm still thinking about all this.

2223
02:06:49,000 --> 02:06:55,001
But what, this, the idea that the maximum
utility is extracted from spectrum

2224
02:06:55,001 --> 02:06:58,001
when everybody is sort of
experiencing maximum interference

2225
02:06:58,001 --> 02:07:02,000
kind of taught me something.

2226
02:07:03,000 --> 02:07:07,001
And I think the FCC, you know, the
micromanagement of the spectrum, which is more

2227
02:07:07,001 --> 02:07:12,000
or less what the FCC has been doing for
most of the century, they think they

2228
02:07:12,000 --> 02:07:16,000
started in the 1920s, was the right public
policy given that the radio technology

2229
02:07:16,000 --> 02:07:21,000
we had at the time was stuff
that was linear modulation

2230
02:07:21,000 --> 02:07:22,001
and linear detection basically.

2231
02:07:22,001 --> 02:07:25,000
It's AM, linear modulation,
even single side band.

2232
02:07:26,001 --> 02:07:30,001
But now that we have computers, and,
you know, there's, there's more law, more is

2233
02:07:30,001 --> 02:07:32,001
law thing,
where every 18 months for the same money you

2234
02:07:32,001 --> 02:07:34,000
can buy something that's twice as fast.

2235
02:07:34,000 --> 02:07:38,000
You know, we wait maybe 15 more years
and the computer's gonna be a thousand

2236
02:07:38,000 --> 02:07:41,000
times faster. And that's just a computer.

2237
02:07:41,000 --> 02:07:44,001
Well, even with custom VLSI, the amount
of processing is gonna be able to do to

2238
02:07:44,001 --> 02:07:48,001
go from wide, to take wide band
signals and like figure out, you

2239
02:07:48,001 --> 02:07:50,000
know, who's trying to send you what in them.

2240
02:07:51,001 --> 02:07:57,000
Makes me think that for the future,
where we should be headed for public policy

2241
02:07:57,000 --> 02:07:59,001
for spectrum allocation is deregulation.

2242
02:07:59,001 --> 02:08:06,000
And what I mean by that is,
basically remove rules so that you can transmit

2243
02:08:06,000 --> 02:08:09,000
whatever you want, whenever you want,
more or less without any

2244
02:08:09,000 --> 02:08:10,001
rules. And maybe we need some power caps.

2245
02:08:11,001 --> 02:08:14,001
That goes against your power control because...

2246
02:08:15,001 --> 02:08:18,001
Actually, I mean,
I think you build power control in the

2247
02:08:18,001 --> 02:08:20,001
system. Yeah, there's a problem there.

2248
02:08:20,001 --> 02:08:22,000
And I was like,
they still would have to cooperate.

2249
02:08:23,001 --> 02:08:27,000
No, you could just go to another band.
There's a lot of 2 GHz wide

2250
02:08:27,000 --> 02:08:28,001
bands between 0 and 100 GHz.

2251
02:08:29,001 --> 02:08:30,001
So I actually think...

2252
02:08:31,000 --> 02:08:34,000
It's regulated so your neighbor
could be occupying 17 of them.

2253
02:08:35,000 --> 02:08:41,001
But then there's 93, you know,
83 other GHz wide spectrum. And I think perhaps

2254
02:08:41,001 --> 02:08:46,001
maybe it only makes sense to deregulate
the spectrum above maybe a GHz. Maybe.

2255
02:08:48,001 --> 02:08:53,000
But actually, I don't know.
I have a suggested target rate for... Why say

2256
02:08:53,000 --> 02:08:57,001
deregulate? I mean,
basically get rid of any federal licensing or

2257
02:08:57,001 --> 02:08:59,001
type acceptance rules for regulation.

2258
02:09:00,001 --> 02:09:03,001
And, you know, every time you double the
frequency, you get sort of different

2259
02:09:03,001 --> 02:09:06,001
propagation characteristics. So different
pieces of the spectrum are useful for

2260
02:09:06,001 --> 02:09:10,000
different things.
Even above a GHz, you get different properties.

2261
02:09:10,001 --> 02:09:13,000
And so you might want to build a system
here or you might want to build a system

2262
02:09:13,000 --> 02:09:15,000
there. But I think the sort of...

2263
02:09:16,000 --> 02:09:19,001
You know, where the technology sort of
could take over and we wouldn't need rules

2264
02:09:19,001 --> 02:09:24,000
anymore for the spectrum is that we could do 2%
in each octave each year

2265
02:09:24,000 --> 02:09:26,000
beginning in the year 2000, which would nearly

2266
02:09:26,000 --> 02:09:28,000
complete deregulation by the year 2050.

2267
02:09:28,001 --> 02:09:33,000
That's, I think, my intuition sort of
says would be sort of the right rate at

2268
02:09:33,000 --> 02:09:35,000
which to deregulate for public policy reasons.

2269
02:09:35,000 --> 02:09:39,000
Now, what about amateur radio?
I still want to be able to do sideband and

2270
02:09:39,000 --> 02:09:40,001
see if I can, you know, hear that guy in New

2271
02:09:40,001 --> 02:09:43,000
Zealand who's only transmitting with 5 watts.

2272
02:09:43,001 --> 02:09:48,000
Well, that's shortwave. I've already said that.
But actually, I think it might

2273
02:09:48,000 --> 02:09:55,000
make sense to keep some handband kind of
things around and call them the public

2274
02:09:55,000 --> 02:09:57,000
parts of the spectrum.

2275
02:09:57,000 --> 02:10:01,001
The place where you can actually go play
with what we think of as weak signal or

2276
02:10:01,001 --> 02:10:06,001
linear modulation techniques. Because
there's an important educational aspect to

2277
02:10:06,001 --> 02:10:09,001
actually, you know, transmitting and
playing with the receiver and being able to

2278
02:10:09,001 --> 02:10:12,000
run around with it and doing things
like I was doing, which was Fox-Honing.

2279
02:10:13,000 --> 02:10:18,000
You could Fox-Honed in a spread spectrum
world as well. But I don't think I ever

2280
02:10:18,000 --> 02:10:20,001
would have done this thesis if I wasn't
actually running around playing with

2281
02:10:20,001 --> 02:10:24,000
radios for fun because I learned a lot there.

2282
02:10:24,000 --> 02:10:28,000
And I think it'd be nice to have some
spectrum to play with for learning

2283
02:10:28,000 --> 02:10:31,000
purposes.
You can learn with wideband systems as

2284
02:10:31,000 --> 02:10:32,001
well, even with everybody else out there.

2285
02:10:32,001 --> 02:10:35,001
But there's something nice about actually
being able to turn a receiver on and

2286
02:10:35,001 --> 02:10:38,000
hear the thermal noise floor. And it'd be nice

2287
02:10:38,000 --> 02:10:39,001
if there were some places in the spectrum.

2288
02:10:40,000 --> 02:10:43,001
Now, the other places in the spectrum which
we can never deregulate are the radio

2289
02:10:43,001 --> 02:10:47,001
astronomy people. If they want to continue
to do earth-based radio astronomy or

2290
02:10:47,001 --> 02:10:49,001
they could go build a base
on the far side of the moon.

2291
02:10:50,001 --> 02:10:55,000
But now I'm getting really wacko and that's
really all I had to say today. And I

2292
02:10:55,000 --> 02:10:57,000
can't...

2293
02:10:57,000 --> 02:10:57,001
Thank you.

2294
02:11:01,000 --> 02:11:03,001
And while I talked longer than
I thought I was going to.

2295
02:11:11,001 --> 02:11:13,001
Build a base on the far side of the moon? Yeah.

2296
02:11:15,001 --> 02:11:17,001
So thank you, Greg, for inviting me to do this.

2297
02:11:18,000 --> 02:11:18,001
Thank you.